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 "../include/lustre_lite.h"
42 #include "llite_internal.h"
43 #include "vvp_internal.h"
45 /*****************************************************************************
47 * Vvp device and device type functions.
52 * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
53 * "llite_" (var. "ll_") prefix.
56 static struct kmem_cache *ll_thread_kmem;
57 struct kmem_cache *vvp_lock_kmem;
58 struct kmem_cache *vvp_object_kmem;
59 struct kmem_cache *vvp_req_kmem;
60 static struct kmem_cache *vvp_session_kmem;
61 static struct kmem_cache *vvp_thread_kmem;
63 static struct lu_kmem_descr vvp_caches[] = {
65 .ckd_cache = &ll_thread_kmem,
66 .ckd_name = "ll_thread_kmem",
67 .ckd_size = sizeof(struct ll_thread_info),
70 .ckd_cache = &vvp_lock_kmem,
71 .ckd_name = "vvp_lock_kmem",
72 .ckd_size = sizeof(struct vvp_lock),
75 .ckd_cache = &vvp_object_kmem,
76 .ckd_name = "vvp_object_kmem",
77 .ckd_size = sizeof(struct vvp_object),
80 .ckd_cache = &vvp_req_kmem,
81 .ckd_name = "vvp_req_kmem",
82 .ckd_size = sizeof(struct vvp_req),
85 .ckd_cache = &vvp_session_kmem,
86 .ckd_name = "vvp_session_kmem",
87 .ckd_size = sizeof(struct vvp_session)
90 .ckd_cache = &vvp_thread_kmem,
91 .ckd_name = "vvp_thread_kmem",
92 .ckd_size = sizeof(struct vvp_thread_info),
99 static void *ll_thread_key_init(const struct lu_context *ctx,
100 struct lu_context_key *key)
102 struct vvp_thread_info *info;
104 info = kmem_cache_zalloc(ll_thread_kmem, GFP_NOFS);
106 info = ERR_PTR(-ENOMEM);
110 static void ll_thread_key_fini(const struct lu_context *ctx,
111 struct lu_context_key *key, void *data)
113 struct vvp_thread_info *info = data;
115 kmem_cache_free(ll_thread_kmem, info);
118 struct lu_context_key ll_thread_key = {
119 .lct_tags = LCT_CL_THREAD,
120 .lct_init = ll_thread_key_init,
121 .lct_fini = ll_thread_key_fini
124 static void *vvp_session_key_init(const struct lu_context *ctx,
125 struct lu_context_key *key)
127 struct vvp_session *session;
129 session = kmem_cache_zalloc(vvp_session_kmem, GFP_NOFS);
131 session = ERR_PTR(-ENOMEM);
135 static void vvp_session_key_fini(const struct lu_context *ctx,
136 struct lu_context_key *key, void *data)
138 struct vvp_session *session = data;
140 kmem_cache_free(vvp_session_kmem, session);
143 struct lu_context_key vvp_session_key = {
144 .lct_tags = LCT_SESSION,
145 .lct_init = vvp_session_key_init,
146 .lct_fini = vvp_session_key_fini
149 static void *vvp_thread_key_init(const struct lu_context *ctx,
150 struct lu_context_key *key)
152 struct vvp_thread_info *vti;
154 vti = kmem_cache_zalloc(vvp_thread_kmem, GFP_NOFS);
156 vti = ERR_PTR(-ENOMEM);
160 static void vvp_thread_key_fini(const struct lu_context *ctx,
161 struct lu_context_key *key, void *data)
163 struct vvp_thread_info *vti = data;
165 kmem_cache_free(vvp_thread_kmem, vti);
168 struct lu_context_key vvp_thread_key = {
169 .lct_tags = LCT_CL_THREAD,
170 .lct_init = vvp_thread_key_init,
171 .lct_fini = vvp_thread_key_fini
174 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
175 LU_TYPE_INIT_FINI(vvp, &vvp_thread_key, &ll_thread_key, &vvp_session_key);
177 static const struct lu_device_operations vvp_lu_ops = {
178 .ldo_object_alloc = vvp_object_alloc
181 static const struct cl_device_operations vvp_cl_ops = {
182 .cdo_req_init = vvp_req_init
185 static struct lu_device *vvp_device_free(const struct lu_env *env,
188 struct vvp_device *vdv = lu2vvp_dev(d);
189 struct cl_site *site = lu2cl_site(d->ld_site);
190 struct lu_device *next = cl2lu_dev(vdv->vdv_next);
196 cl_device_fini(lu2cl_dev(d));
201 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
202 struct lu_device_type *t,
203 struct lustre_cfg *cfg)
205 struct vvp_device *vdv;
206 struct lu_device *lud;
207 struct cl_site *site;
210 vdv = kzalloc(sizeof(*vdv), GFP_NOFS);
212 return ERR_PTR(-ENOMEM);
214 lud = &vdv->vdv_cl.cd_lu_dev;
215 cl_device_init(&vdv->vdv_cl, t);
216 vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
217 vdv->vdv_cl.cd_ops = &vvp_cl_ops;
219 site = kzalloc(sizeof(*site), GFP_NOFS);
221 rc = cl_site_init(site, &vdv->vdv_cl);
223 rc = lu_site_init_finish(&site->cs_lu);
225 LASSERT(!lud->ld_site);
226 CERROR("Cannot init lu_site, rc %d.\n", rc);
233 vvp_device_free(env, lud);
239 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
240 const char *name, struct lu_device *next)
242 struct vvp_device *vdv;
246 vdv->vdv_next = lu2cl_dev(next);
248 LASSERT(d->ld_site && next->ld_type);
249 next->ld_site = d->ld_site;
250 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
251 next->ld_type->ldt_name,
255 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
260 static struct lu_device *vvp_device_fini(const struct lu_env *env,
263 return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
266 static const struct lu_device_type_operations vvp_device_type_ops = {
267 .ldto_init = vvp_type_init,
268 .ldto_fini = vvp_type_fini,
270 .ldto_start = vvp_type_start,
271 .ldto_stop = vvp_type_stop,
273 .ldto_device_alloc = vvp_device_alloc,
274 .ldto_device_free = vvp_device_free,
275 .ldto_device_init = vvp_device_init,
276 .ldto_device_fini = vvp_device_fini,
279 struct lu_device_type vvp_device_type = {
280 .ldt_tags = LU_DEVICE_CL,
281 .ldt_name = LUSTRE_VVP_NAME,
282 .ldt_ops = &vvp_device_type_ops,
283 .ldt_ctx_tags = LCT_CL_THREAD
287 * A mutex serializing calls to vvp_inode_fini() under extreme memory
288 * pressure, when environments cannot be allocated.
290 int vvp_global_init(void)
294 rc = lu_kmem_init(vvp_caches);
298 rc = lu_device_type_init(&vvp_device_type);
305 lu_kmem_fini(vvp_caches);
310 void vvp_global_fini(void)
312 lu_device_type_fini(&vvp_device_type);
313 lu_kmem_fini(vvp_caches);
316 /*****************************************************************************
318 * mirror obd-devices into cl devices.
322 int cl_sb_init(struct super_block *sb)
324 struct ll_sb_info *sbi;
325 struct cl_device *cl;
331 env = cl_env_get(&refcheck);
333 cl = cl_type_setup(env, NULL, &vvp_device_type,
334 sbi->ll_dt_exp->exp_obd->obd_lu_dev);
336 cl2vvp_dev(cl)->vdv_sb = sb;
338 sbi->ll_site = cl2lu_dev(cl)->ld_site;
340 cl_env_put(env, &refcheck);
347 int cl_sb_fini(struct super_block *sb)
349 struct ll_sb_info *sbi;
351 struct cl_device *cld;
356 env = cl_env_get(&refcheck);
361 cl_stack_fini(env, cld);
365 cl_env_put(env, &refcheck);
368 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
369 result = PTR_ERR(env);
372 * If mount failed (sbi->ll_cl == NULL), and this there are no other
373 * mounts, stop device types manually (this usually happens
374 * automatically when last device is destroyed).
380 /****************************************************************************
382 * debugfs/lustre/llite/$MNT/dump_page_cache
384 ****************************************************************************/
387 * To represent contents of a page cache as a byte stream, following
388 * information if encoded in 64bit offset:
390 * - file hash bucket in lu_site::ls_hash[] 28bits
392 * - how far file is from bucket head 4bits
394 * - page index 32bits
396 * First two data identify a file in the cache uniquely.
399 #define PGC_OBJ_SHIFT (32 + 4)
400 #define PGC_DEPTH_SHIFT (32)
402 struct vvp_pgcache_id {
408 struct lu_object_header *vpi_obj;
411 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
413 CLASSERT(sizeof(pos) == sizeof(__u64));
415 id->vpi_index = pos & 0xffffffff;
416 id->vpi_depth = (pos >> PGC_DEPTH_SHIFT) & 0xf;
417 id->vpi_bucket = (unsigned long long)pos >> PGC_OBJ_SHIFT;
420 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
423 ((__u64)id->vpi_index) |
424 ((__u64)id->vpi_depth << PGC_DEPTH_SHIFT) |
425 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
428 static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
429 struct hlist_node *hnode, void *data)
431 struct vvp_pgcache_id *id = data;
432 struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
434 if (id->vpi_curdep-- > 0)
435 return 0; /* continue */
437 if (lu_object_is_dying(hdr))
440 cfs_hash_get(hs, hnode);
445 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
446 struct lu_device *dev,
447 struct vvp_pgcache_id *id)
449 LASSERT(lu_device_is_cl(dev));
451 id->vpi_depth &= 0xf;
453 id->vpi_curdep = id->vpi_depth;
455 cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
456 vvp_pgcache_obj_get, id);
458 struct lu_object *lu_obj;
460 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
462 lu_object_ref_add(lu_obj, "dump", current);
463 return lu2cl(lu_obj);
465 lu_object_put(env, lu_object_top(id->vpi_obj));
467 } else if (id->vpi_curdep > 0) {
473 static loff_t vvp_pgcache_find(const struct lu_env *env,
474 struct lu_device *dev, loff_t pos)
476 struct cl_object *clob;
477 struct lu_site *site;
478 struct vvp_pgcache_id id;
481 vvp_pgcache_id_unpack(pos, &id);
484 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
486 clob = vvp_pgcache_obj(env, dev, &id);
488 struct inode *inode = vvp_object_inode(clob);
492 nr = find_get_pages_contig(inode->i_mapping,
493 id.vpi_index, 1, &vmpage);
495 id.vpi_index = vmpage->index;
496 /* Cant support over 16T file */
497 nr = !(vmpage->index > 0xffffffff);
501 lu_object_ref_del(&clob->co_lu, "dump", current);
502 cl_object_put(env, clob);
504 return vvp_pgcache_id_pack(&id);
506 /* to the next object. */
509 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
515 #define seq_page_flag(seq, page, flag, has_flags) do { \
516 if (test_bit(PG_##flag, &(page)->flags)) { \
517 seq_printf(seq, "%s"#flag, has_flags ? "|" : ""); \
522 static void vvp_pgcache_page_show(const struct lu_env *env,
523 struct seq_file *seq, struct cl_page *page)
525 struct vvp_page *vpg;
529 vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
530 vmpage = vpg->vpg_page;
531 seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
535 vpg->vpg_write_queued ? "wq" : "- ",
536 vpg->vpg_defer_uptodate ? "du" : "- ",
537 PageWriteback(vmpage) ? "wb" : "-",
538 vmpage, PFID(ll_inode2fid(vmpage->mapping->host)),
539 vmpage->mapping->host, vmpage->index,
542 seq_page_flag(seq, vmpage, locked, has_flags);
543 seq_page_flag(seq, vmpage, error, has_flags);
544 seq_page_flag(seq, vmpage, referenced, has_flags);
545 seq_page_flag(seq, vmpage, uptodate, has_flags);
546 seq_page_flag(seq, vmpage, dirty, has_flags);
547 seq_page_flag(seq, vmpage, writeback, has_flags);
548 seq_printf(seq, "%s]\n", has_flags ? "" : "-");
551 static int vvp_pgcache_show(struct seq_file *f, void *v)
554 struct ll_sb_info *sbi;
555 struct cl_object *clob;
557 struct vvp_pgcache_id id;
561 env = cl_env_get(&refcheck);
564 vvp_pgcache_id_unpack(pos, &id);
566 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
568 struct inode *inode = vvp_object_inode(clob);
569 struct cl_page *page = NULL;
572 result = find_get_pages_contig(inode->i_mapping,
577 page = cl_vmpage_page(vmpage, clob);
582 seq_printf(f, "%8x@" DFID ": ", id.vpi_index,
583 PFID(lu_object_fid(&clob->co_lu)));
585 vvp_pgcache_page_show(env, f, page);
586 cl_page_put(env, page);
588 seq_puts(f, "missing\n");
590 lu_object_ref_del(&clob->co_lu, "dump", current);
591 cl_object_put(env, clob);
593 seq_printf(f, "%llx missing\n", pos);
595 cl_env_put(env, &refcheck);
598 result = PTR_ERR(env);
603 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
605 struct ll_sb_info *sbi;
611 env = cl_env_get(&refcheck);
614 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
615 pos = ERR_PTR(-EFBIG);
617 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
622 cl_env_put(env, &refcheck);
627 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
629 struct ll_sb_info *sbi;
633 env = cl_env_get(&refcheck);
636 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
639 cl_env_put(env, &refcheck);
644 static void vvp_pgcache_stop(struct seq_file *f, void *v)
649 static const struct seq_operations vvp_pgcache_ops = {
650 .start = vvp_pgcache_start,
651 .next = vvp_pgcache_next,
652 .stop = vvp_pgcache_stop,
653 .show = vvp_pgcache_show
656 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
658 struct seq_file *seq;
661 rc = seq_open(filp, &vvp_pgcache_ops);
665 seq = filp->private_data;
666 seq->private = inode->i_private;
671 const struct file_operations vvp_dump_pgcache_file_ops = {
672 .owner = THIS_MODULE,
673 .open = vvp_dump_pgcache_seq_open,
676 .release = seq_release,