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) 2012, 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_device and cl_device_type implementation for VVP layer.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
42 #define DEBUG_SUBSYSTEM S_LLITE
46 #include "llite_internal.h"
47 #include "vvp_internal.h"
49 /*****************************************************************************
51 * Vvp device and device type functions.
56 * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
57 * "llite_" (var. "ll_") prefix.
60 struct kmem_cache *vvp_lock_kmem;
61 struct kmem_cache *vvp_object_kmem;
62 static struct kmem_cache *vvp_thread_kmem;
63 static struct kmem_cache *vvp_session_kmem;
64 static struct lu_kmem_descr vvp_caches[] = {
66 .ckd_cache = &vvp_lock_kmem,
67 .ckd_name = "vvp_lock_kmem",
68 .ckd_size = sizeof(struct vvp_lock),
71 .ckd_cache = &vvp_object_kmem,
72 .ckd_name = "vvp_object_kmem",
73 .ckd_size = sizeof(struct vvp_object),
76 .ckd_cache = &vvp_thread_kmem,
77 .ckd_name = "vvp_thread_kmem",
78 .ckd_size = sizeof (struct vvp_thread_info),
81 .ckd_cache = &vvp_session_kmem,
82 .ckd_name = "vvp_session_kmem",
83 .ckd_size = sizeof (struct vvp_session)
90 static void *vvp_key_init(const struct lu_context *ctx,
91 struct lu_context_key *key)
93 struct vvp_thread_info *info;
95 OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, GFP_NOFS);
97 info = ERR_PTR(-ENOMEM);
101 static void vvp_key_fini(const struct lu_context *ctx,
102 struct lu_context_key *key, void *data)
104 struct vvp_thread_info *info = data;
105 OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
108 static void *vvp_session_key_init(const struct lu_context *ctx,
109 struct lu_context_key *key)
111 struct vvp_session *session;
113 OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
115 session = ERR_PTR(-ENOMEM);
119 static void vvp_session_key_fini(const struct lu_context *ctx,
120 struct lu_context_key *key, void *data)
122 struct vvp_session *session = data;
123 OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
127 struct lu_context_key vvp_key = {
128 .lct_tags = LCT_CL_THREAD,
129 .lct_init = vvp_key_init,
130 .lct_fini = vvp_key_fini
133 struct lu_context_key vvp_session_key = {
134 .lct_tags = LCT_SESSION,
135 .lct_init = vvp_session_key_init,
136 .lct_fini = vvp_session_key_fini
139 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
140 LU_TYPE_INIT_FINI(vvp, &ccc_key, &vvp_key, &vvp_session_key);
142 static const struct lu_device_operations vvp_lu_ops = {
143 .ldo_object_alloc = vvp_object_alloc
146 static const struct cl_device_operations vvp_cl_ops = {
147 .cdo_req_init = ccc_req_init
150 static struct lu_device *vvp_device_free(const struct lu_env *env,
153 struct vvp_device *vdv = lu2vvp_dev(d);
154 struct cl_site *site = lu2cl_site(d->ld_site);
155 struct lu_device *next = cl2lu_dev(vdv->vdv_next);
157 if (d->ld_site != NULL) {
162 cl_device_fini(lu2cl_dev(d));
167 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
168 struct lu_device_type *t,
169 struct lustre_cfg *cfg)
171 struct vvp_device *vdv;
172 struct lu_device *lud;
173 struct cl_site *site;
179 RETURN(ERR_PTR(-ENOMEM));
181 lud = &vdv->vdv_cl.cd_lu_dev;
182 cl_device_init(&vdv->vdv_cl, t);
183 vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
184 vdv->vdv_cl.cd_ops = &vvp_cl_ops;
188 rc = cl_site_init(site, &vdv->vdv_cl);
190 rc = lu_site_init_finish(&site->cs_lu);
192 LASSERT(lud->ld_site == NULL);
193 CERROR("Cannot init lu_site, rc %d.\n", rc);
199 vvp_device_free(env, lud);
205 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
206 const char *name, struct lu_device *next)
208 struct vvp_device *vdv;
213 vdv->vdv_next = lu2cl_dev(next);
215 LASSERT(d->ld_site != NULL && next->ld_type != NULL);
216 next->ld_site = d->ld_site;
217 rc = next->ld_type->ldt_ops->ldto_device_init(
218 env, next, next->ld_type->ldt_name, NULL);
221 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
226 static struct lu_device *vvp_device_fini(const struct lu_env *env,
229 return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
232 static const struct lu_device_type_operations vvp_device_type_ops = {
233 .ldto_init = vvp_type_init,
234 .ldto_fini = vvp_type_fini,
236 .ldto_start = vvp_type_start,
237 .ldto_stop = vvp_type_stop,
239 .ldto_device_alloc = vvp_device_alloc,
240 .ldto_device_free = vvp_device_free,
241 .ldto_device_init = vvp_device_init,
242 .ldto_device_fini = vvp_device_fini,
245 struct lu_device_type vvp_device_type = {
246 .ldt_tags = LU_DEVICE_CL,
247 .ldt_name = LUSTRE_VVP_NAME,
248 .ldt_ops = &vvp_device_type_ops,
249 .ldt_ctx_tags = LCT_CL_THREAD
253 * A mutex serializing calls to vvp_inode_fini() under extreme memory
254 * pressure, when environments cannot be allocated.
256 int vvp_global_init(void)
260 result = lu_kmem_init(vvp_caches);
262 result = ccc_global_init(&vvp_device_type);
264 lu_kmem_fini(vvp_caches);
269 void vvp_global_fini(void)
271 ccc_global_fini(&vvp_device_type);
272 lu_kmem_fini(vvp_caches);
276 /*****************************************************************************
278 * mirror obd-devices into cl devices.
282 int cl_sb_init(struct super_block *sb)
284 struct ll_sb_info *sbi;
285 struct cl_device *cl;
291 env = cl_env_get(&refcheck);
293 cl = cl_type_setup(env, NULL, &vvp_device_type,
294 sbi->ll_dt_exp->exp_obd->obd_lu_dev);
296 cl2vvp_dev(cl)->vdv_sb = sb;
298 sbi->ll_site = cl2lu_dev(cl)->ld_site;
300 cl_env_put(env, &refcheck);
306 int cl_sb_fini(struct super_block *sb)
308 struct ll_sb_info *sbi;
310 struct cl_device *cld;
316 env = cl_env_get(&refcheck);
321 cl_stack_fini(env, cld);
325 cl_env_put(env, &refcheck);
328 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
329 result = PTR_ERR(env);
335 /****************************************************************************
337 * /proc/fs/lustre/llite/$MNT/dump_page_cache
339 ****************************************************************************/
342 * To represent contents of a page cache as a byte stream, following
343 * information if encoded in 64bit offset:
345 * - file hash bucket in lu_site::ls_hash[] 28bits
347 * - how far file is from bucket head 4bits
349 * - page index 32bits
351 * First two data identify a file in the cache uniquely.
354 #define PGC_OBJ_SHIFT (32 + 4)
355 #define PGC_DEPTH_SHIFT (32)
357 struct vvp_pgcache_id {
363 struct lu_object_header *vpi_obj;
366 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
368 CLASSERT(sizeof(pos) == sizeof(__u64));
370 id->vpi_index = pos & 0xffffffff;
371 id->vpi_depth = (pos >> PGC_DEPTH_SHIFT) & 0xf;
372 id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
375 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
378 ((__u64)id->vpi_index) |
379 ((__u64)id->vpi_depth << PGC_DEPTH_SHIFT) |
380 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
383 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
384 struct hlist_node *hnode, void *data)
386 struct vvp_pgcache_id *id = data;
387 struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
389 if (id->vpi_curdep-- > 0)
390 return 0; /* continue */
392 if (lu_object_is_dying(hdr))
395 cfs_hash_get(hs, hnode);
400 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
401 struct lu_device *dev,
402 struct vvp_pgcache_id *id)
404 LASSERT(lu_device_is_cl(dev));
406 id->vpi_depth &= 0xf;
408 id->vpi_curdep = id->vpi_depth;
410 cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
411 vvp_pgcache_obj_get, id);
412 if (id->vpi_obj != NULL) {
413 struct lu_object *lu_obj;
415 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
416 if (lu_obj != NULL) {
417 lu_object_ref_add(lu_obj, "dump", current);
418 return lu2cl(lu_obj);
420 lu_object_put(env, lu_object_top(id->vpi_obj));
422 } else if (id->vpi_curdep > 0) {
428 static loff_t vvp_pgcache_find(const struct lu_env *env,
429 struct lu_device *dev, loff_t pos)
431 struct cl_object *clob;
432 struct lu_site *site;
433 struct vvp_pgcache_id id;
436 vvp_pgcache_id_unpack(pos, &id);
439 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
441 clob = vvp_pgcache_obj(env, dev, &id);
443 struct inode *inode = vvp_object_inode(clob);
447 nr = find_get_pages_contig(inode->i_mapping,
448 id.vpi_index, 1, &vmpage);
450 id.vpi_index = vmpage->index;
451 /* Cant support over 16T file */
452 nr = !(vmpage->index > 0xffffffff);
453 page_cache_release(vmpage);
456 lu_object_ref_del(&clob->co_lu, "dump", current);
457 cl_object_put(env, clob);
459 return vvp_pgcache_id_pack(&id);
461 /* to the next object. */
464 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
470 #define seq_page_flag(seq, page, flag, has_flags) do { \
471 if (test_bit(PG_##flag, &(page)->flags)) { \
472 seq_printf(seq, "%s"#flag, has_flags ? "|" : ""); \
477 static void vvp_pgcache_page_show(const struct lu_env *env,
478 struct seq_file *seq, struct cl_page *page)
480 struct vvp_page *vpg;
484 vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
485 vmpage = vpg->vpg_page;
486 seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
490 vpg->vpg_write_queued ? "wq" : "- ",
491 vpg->vpg_defer_uptodate ? "du" : "- ",
492 PageWriteback(vmpage) ? "wb" : "-",
494 PFID(ll_inode2fid(vmpage->mapping->host)),
495 vmpage->mapping->host, vmpage->index,
498 seq_page_flag(seq, vmpage, locked, has_flags);
499 seq_page_flag(seq, vmpage, error, has_flags);
500 seq_page_flag(seq, vmpage, referenced, has_flags);
501 seq_page_flag(seq, vmpage, uptodate, has_flags);
502 seq_page_flag(seq, vmpage, dirty, has_flags);
503 seq_page_flag(seq, vmpage, writeback, has_flags);
504 seq_printf(seq, "%s]\n", has_flags ? "" : "-");
507 static int vvp_pgcache_show(struct seq_file *f, void *v)
510 struct ll_sb_info *sbi;
511 struct cl_object *clob;
513 struct vvp_pgcache_id id;
517 env = cl_env_get(&refcheck);
520 vvp_pgcache_id_unpack(pos, &id);
522 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
524 struct inode *inode = vvp_object_inode(clob);
525 struct cl_page *page = NULL;
528 result = find_get_pages_contig(inode->i_mapping,
529 id.vpi_index, 1, &vmpage);
532 page = cl_vmpage_page(vmpage, clob);
535 page_cache_release(vmpage);
538 seq_printf(f, "%8x@"DFID": ", id.vpi_index,
539 PFID(lu_object_fid(&clob->co_lu)));
541 vvp_pgcache_page_show(env, f, page);
542 cl_page_put(env, page);
544 seq_puts(f, "missing\n");
545 lu_object_ref_del(&clob->co_lu, "dump", current);
546 cl_object_put(env, clob);
548 seq_printf(f, "%llx missing\n", pos);
549 cl_env_put(env, &refcheck);
552 result = PTR_ERR(env);
556 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
558 struct ll_sb_info *sbi;
564 env = cl_env_get(&refcheck);
567 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
568 pos = ERR_PTR(-EFBIG);
570 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
575 cl_env_put(env, &refcheck);
580 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
582 struct ll_sb_info *sbi;
586 env = cl_env_get(&refcheck);
589 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
592 cl_env_put(env, &refcheck);
597 static void vvp_pgcache_stop(struct seq_file *f, void *v)
602 static struct seq_operations vvp_pgcache_ops = {
603 .start = vvp_pgcache_start,
604 .next = vvp_pgcache_next,
605 .stop = vvp_pgcache_stop,
606 .show = vvp_pgcache_show
609 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
611 struct ll_sb_info *sbi = PDE_DATA(inode);
612 struct seq_file *seq;
615 result = seq_open(filp, &vvp_pgcache_ops);
617 seq = filp->private_data;
623 const struct file_operations vvp_dump_pgcache_file_ops = {
624 .owner = THIS_MODULE,
625 .open = vvp_dump_pgcache_seq_open,
628 .release = seq_release,