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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * cl_device and cl_device_type implementation for VVP layer.
34 * Author: Nikita Danilov <nikita.danilov@sun.com>
35 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
38 #define DEBUG_SUBSYSTEM S_LLITE
40 #include "../include/obd.h"
41 #include "llite_internal.h"
42 #include "vvp_internal.h"
44 /*****************************************************************************
46 * Vvp device and device type functions.
51 * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
52 * "llite_" (var. "ll_") prefix.
55 static struct kmem_cache *ll_thread_kmem;
56 struct kmem_cache *vvp_lock_kmem;
57 struct kmem_cache *vvp_object_kmem;
58 struct kmem_cache *vvp_req_kmem;
59 static struct kmem_cache *vvp_session_kmem;
60 static struct kmem_cache *vvp_thread_kmem;
62 static struct lu_kmem_descr vvp_caches[] = {
64 .ckd_cache = &ll_thread_kmem,
65 .ckd_name = "ll_thread_kmem",
66 .ckd_size = sizeof(struct ll_thread_info),
69 .ckd_cache = &vvp_lock_kmem,
70 .ckd_name = "vvp_lock_kmem",
71 .ckd_size = sizeof(struct vvp_lock),
74 .ckd_cache = &vvp_object_kmem,
75 .ckd_name = "vvp_object_kmem",
76 .ckd_size = sizeof(struct vvp_object),
79 .ckd_cache = &vvp_req_kmem,
80 .ckd_name = "vvp_req_kmem",
81 .ckd_size = sizeof(struct vvp_req),
84 .ckd_cache = &vvp_session_kmem,
85 .ckd_name = "vvp_session_kmem",
86 .ckd_size = sizeof(struct vvp_session)
89 .ckd_cache = &vvp_thread_kmem,
90 .ckd_name = "vvp_thread_kmem",
91 .ckd_size = sizeof(struct vvp_thread_info),
98 static void *ll_thread_key_init(const struct lu_context *ctx,
99 struct lu_context_key *key)
101 struct vvp_thread_info *info;
103 info = kmem_cache_zalloc(ll_thread_kmem, GFP_NOFS);
105 info = ERR_PTR(-ENOMEM);
109 static void ll_thread_key_fini(const struct lu_context *ctx,
110 struct lu_context_key *key, void *data)
112 struct vvp_thread_info *info = data;
114 kmem_cache_free(ll_thread_kmem, info);
117 struct lu_context_key ll_thread_key = {
118 .lct_tags = LCT_CL_THREAD,
119 .lct_init = ll_thread_key_init,
120 .lct_fini = ll_thread_key_fini
123 static void *vvp_session_key_init(const struct lu_context *ctx,
124 struct lu_context_key *key)
126 struct vvp_session *session;
128 session = kmem_cache_zalloc(vvp_session_kmem, GFP_NOFS);
130 session = ERR_PTR(-ENOMEM);
134 static void vvp_session_key_fini(const struct lu_context *ctx,
135 struct lu_context_key *key, void *data)
137 struct vvp_session *session = data;
139 kmem_cache_free(vvp_session_kmem, session);
142 struct lu_context_key vvp_session_key = {
143 .lct_tags = LCT_SESSION,
144 .lct_init = vvp_session_key_init,
145 .lct_fini = vvp_session_key_fini
148 static void *vvp_thread_key_init(const struct lu_context *ctx,
149 struct lu_context_key *key)
151 struct vvp_thread_info *vti;
153 vti = kmem_cache_zalloc(vvp_thread_kmem, GFP_NOFS);
155 vti = ERR_PTR(-ENOMEM);
159 static void vvp_thread_key_fini(const struct lu_context *ctx,
160 struct lu_context_key *key, void *data)
162 struct vvp_thread_info *vti = data;
164 kmem_cache_free(vvp_thread_kmem, vti);
167 struct lu_context_key vvp_thread_key = {
168 .lct_tags = LCT_CL_THREAD,
169 .lct_init = vvp_thread_key_init,
170 .lct_fini = vvp_thread_key_fini
173 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
174 LU_TYPE_INIT_FINI(vvp, &vvp_thread_key, &ll_thread_key, &vvp_session_key);
176 static const struct lu_device_operations vvp_lu_ops = {
177 .ldo_object_alloc = vvp_object_alloc
180 static const struct cl_device_operations vvp_cl_ops = {
181 .cdo_req_init = vvp_req_init
184 static struct lu_device *vvp_device_free(const struct lu_env *env,
187 struct vvp_device *vdv = lu2vvp_dev(d);
188 struct cl_site *site = lu2cl_site(d->ld_site);
189 struct lu_device *next = cl2lu_dev(vdv->vdv_next);
195 cl_device_fini(lu2cl_dev(d));
200 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
201 struct lu_device_type *t,
202 struct lustre_cfg *cfg)
204 struct vvp_device *vdv;
205 struct lu_device *lud;
206 struct cl_site *site;
209 vdv = kzalloc(sizeof(*vdv), GFP_NOFS);
211 return ERR_PTR(-ENOMEM);
213 lud = &vdv->vdv_cl.cd_lu_dev;
214 cl_device_init(&vdv->vdv_cl, t);
215 vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
216 vdv->vdv_cl.cd_ops = &vvp_cl_ops;
218 site = kzalloc(sizeof(*site), GFP_NOFS);
220 rc = cl_site_init(site, &vdv->vdv_cl);
222 rc = lu_site_init_finish(&site->cs_lu);
224 LASSERT(!lud->ld_site);
225 CERROR("Cannot init lu_site, rc %d.\n", rc);
232 vvp_device_free(env, lud);
238 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
239 const char *name, struct lu_device *next)
241 struct vvp_device *vdv;
245 vdv->vdv_next = lu2cl_dev(next);
247 LASSERT(d->ld_site && next->ld_type);
248 next->ld_site = d->ld_site;
249 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
250 next->ld_type->ldt_name,
254 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
259 static struct lu_device *vvp_device_fini(const struct lu_env *env,
262 return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
265 static const struct lu_device_type_operations vvp_device_type_ops = {
266 .ldto_init = vvp_type_init,
267 .ldto_fini = vvp_type_fini,
269 .ldto_start = vvp_type_start,
270 .ldto_stop = vvp_type_stop,
272 .ldto_device_alloc = vvp_device_alloc,
273 .ldto_device_free = vvp_device_free,
274 .ldto_device_init = vvp_device_init,
275 .ldto_device_fini = vvp_device_fini,
278 struct lu_device_type vvp_device_type = {
279 .ldt_tags = LU_DEVICE_CL,
280 .ldt_name = LUSTRE_VVP_NAME,
281 .ldt_ops = &vvp_device_type_ops,
282 .ldt_ctx_tags = LCT_CL_THREAD
286 * A mutex serializing calls to vvp_inode_fini() under extreme memory
287 * pressure, when environments cannot be allocated.
289 int vvp_global_init(void)
293 rc = lu_kmem_init(vvp_caches);
297 rc = lu_device_type_init(&vvp_device_type);
304 lu_kmem_fini(vvp_caches);
309 void vvp_global_fini(void)
311 lu_device_type_fini(&vvp_device_type);
312 lu_kmem_fini(vvp_caches);
315 /*****************************************************************************
317 * mirror obd-devices into cl devices.
321 int cl_sb_init(struct super_block *sb)
323 struct ll_sb_info *sbi;
324 struct cl_device *cl;
330 env = cl_env_get(&refcheck);
332 cl = cl_type_setup(env, NULL, &vvp_device_type,
333 sbi->ll_dt_exp->exp_obd->obd_lu_dev);
335 cl2vvp_dev(cl)->vdv_sb = sb;
337 sbi->ll_site = cl2lu_dev(cl)->ld_site;
339 cl_env_put(env, &refcheck);
346 int cl_sb_fini(struct super_block *sb)
348 struct ll_sb_info *sbi;
350 struct cl_device *cld;
355 env = cl_env_get(&refcheck);
360 cl_stack_fini(env, cld);
364 cl_env_put(env, &refcheck);
367 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
368 result = PTR_ERR(env);
373 /****************************************************************************
375 * debugfs/lustre/llite/$MNT/dump_page_cache
377 ****************************************************************************/
380 * To represent contents of a page cache as a byte stream, following
381 * information if encoded in 64bit offset:
383 * - file hash bucket in lu_site::ls_hash[] 28bits
385 * - how far file is from bucket head 4bits
387 * - page index 32bits
389 * First two data identify a file in the cache uniquely.
392 #define PGC_OBJ_SHIFT (32 + 4)
393 #define PGC_DEPTH_SHIFT (32)
395 struct vvp_pgcache_id {
401 struct lu_object_header *vpi_obj;
404 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
406 CLASSERT(sizeof(pos) == sizeof(__u64));
408 id->vpi_index = pos & 0xffffffff;
409 id->vpi_depth = (pos >> PGC_DEPTH_SHIFT) & 0xf;
410 id->vpi_bucket = (unsigned long long)pos >> PGC_OBJ_SHIFT;
413 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
416 ((__u64)id->vpi_index) |
417 ((__u64)id->vpi_depth << PGC_DEPTH_SHIFT) |
418 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
421 static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
422 struct hlist_node *hnode, void *data)
424 struct vvp_pgcache_id *id = data;
425 struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
427 if (id->vpi_curdep-- > 0)
428 return 0; /* continue */
430 if (lu_object_is_dying(hdr))
433 cfs_hash_get(hs, hnode);
438 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
439 struct lu_device *dev,
440 struct vvp_pgcache_id *id)
442 LASSERT(lu_device_is_cl(dev));
444 id->vpi_depth &= 0xf;
446 id->vpi_curdep = id->vpi_depth;
448 cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
449 vvp_pgcache_obj_get, id);
451 struct lu_object *lu_obj;
453 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
455 lu_object_ref_add(lu_obj, "dump", current);
456 return lu2cl(lu_obj);
458 lu_object_put(env, lu_object_top(id->vpi_obj));
460 } else if (id->vpi_curdep > 0) {
466 static loff_t vvp_pgcache_find(const struct lu_env *env,
467 struct lu_device *dev, loff_t pos)
469 struct cl_object *clob;
470 struct lu_site *site;
471 struct vvp_pgcache_id id;
474 vvp_pgcache_id_unpack(pos, &id);
477 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
479 clob = vvp_pgcache_obj(env, dev, &id);
481 struct inode *inode = vvp_object_inode(clob);
485 nr = find_get_pages_contig(inode->i_mapping,
486 id.vpi_index, 1, &vmpage);
488 id.vpi_index = vmpage->index;
489 /* Cant support over 16T file */
490 nr = !(vmpage->index > 0xffffffff);
494 lu_object_ref_del(&clob->co_lu, "dump", current);
495 cl_object_put(env, clob);
497 return vvp_pgcache_id_pack(&id);
499 /* to the next object. */
502 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
508 #define seq_page_flag(seq, page, flag, has_flags) do { \
509 if (test_bit(PG_##flag, &(page)->flags)) { \
510 seq_printf(seq, "%s"#flag, has_flags ? "|" : ""); \
515 static void vvp_pgcache_page_show(const struct lu_env *env,
516 struct seq_file *seq, struct cl_page *page)
518 struct vvp_page *vpg;
522 vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
523 vmpage = vpg->vpg_page;
524 seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
528 vpg->vpg_write_queued ? "wq" : "- ",
529 vpg->vpg_defer_uptodate ? "du" : "- ",
530 PageWriteback(vmpage) ? "wb" : "-",
531 vmpage, PFID(ll_inode2fid(vmpage->mapping->host)),
532 vmpage->mapping->host, vmpage->index,
535 seq_page_flag(seq, vmpage, locked, has_flags);
536 seq_page_flag(seq, vmpage, error, has_flags);
537 seq_page_flag(seq, vmpage, referenced, has_flags);
538 seq_page_flag(seq, vmpage, uptodate, has_flags);
539 seq_page_flag(seq, vmpage, dirty, has_flags);
540 seq_page_flag(seq, vmpage, writeback, has_flags);
541 seq_printf(seq, "%s]\n", has_flags ? "" : "-");
544 static int vvp_pgcache_show(struct seq_file *f, void *v)
547 struct ll_sb_info *sbi;
548 struct cl_object *clob;
550 struct vvp_pgcache_id id;
554 env = cl_env_get(&refcheck);
557 vvp_pgcache_id_unpack(pos, &id);
559 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
561 struct inode *inode = vvp_object_inode(clob);
562 struct cl_page *page = NULL;
565 result = find_get_pages_contig(inode->i_mapping,
570 page = cl_vmpage_page(vmpage, clob);
575 seq_printf(f, "%8x@" DFID ": ", id.vpi_index,
576 PFID(lu_object_fid(&clob->co_lu)));
578 vvp_pgcache_page_show(env, f, page);
579 cl_page_put(env, page);
581 seq_puts(f, "missing\n");
583 lu_object_ref_del(&clob->co_lu, "dump", current);
584 cl_object_put(env, clob);
586 seq_printf(f, "%llx missing\n", pos);
588 cl_env_put(env, &refcheck);
591 result = PTR_ERR(env);
596 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
598 struct ll_sb_info *sbi;
604 env = cl_env_get(&refcheck);
607 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
608 pos = ERR_PTR(-EFBIG);
610 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
615 cl_env_put(env, &refcheck);
620 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
622 struct ll_sb_info *sbi;
626 env = cl_env_get(&refcheck);
629 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
632 cl_env_put(env, &refcheck);
637 static void vvp_pgcache_stop(struct seq_file *f, void *v)
642 static const struct seq_operations vvp_pgcache_ops = {
643 .start = vvp_pgcache_start,
644 .next = vvp_pgcache_next,
645 .stop = vvp_pgcache_stop,
646 .show = vvp_pgcache_show
649 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
651 struct seq_file *seq;
654 rc = seq_open(filp, &vvp_pgcache_ops);
658 seq = filp->private_data;
659 seq->private = inode->i_private;
664 const struct file_operations vvp_dump_pgcache_file_ops = {
665 .owner = THIS_MODULE,
666 .open = vvp_dump_pgcache_seq_open,
669 .release = seq_release,