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, 2015, 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
44 #include "../include/obd.h"
45 #include "../include/lustre_lite.h"
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 info = kmem_cache_zalloc(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;
106 kmem_cache_free(vvp_thread_kmem, info);
109 static void *vvp_session_key_init(const struct lu_context *ctx,
110 struct lu_context_key *key)
112 struct vvp_session *session;
114 session = kmem_cache_zalloc(vvp_session_kmem, GFP_NOFS);
116 session = ERR_PTR(-ENOMEM);
120 static void vvp_session_key_fini(const struct lu_context *ctx,
121 struct lu_context_key *key, void *data)
123 struct vvp_session *session = data;
125 kmem_cache_free(vvp_session_kmem, session);
128 struct lu_context_key vvp_key = {
129 .lct_tags = LCT_CL_THREAD,
130 .lct_init = vvp_key_init,
131 .lct_fini = vvp_key_fini
134 struct lu_context_key vvp_session_key = {
135 .lct_tags = LCT_SESSION,
136 .lct_init = vvp_session_key_init,
137 .lct_fini = vvp_session_key_fini
140 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
141 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
143 static const struct lu_device_operations vvp_lu_ops = {
144 .ldo_object_alloc = vvp_object_alloc
147 static const struct cl_device_operations vvp_cl_ops = {
148 .cdo_req_init = ccc_req_init
151 static struct lu_device *vvp_device_free(const struct lu_env *env,
154 struct vvp_device *vdv = lu2vvp_dev(d);
155 struct cl_site *site = lu2cl_site(d->ld_site);
156 struct lu_device *next = cl2lu_dev(vdv->vdv_next);
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;
176 vdv = kzalloc(sizeof(*vdv), GFP_NOFS);
178 return ERR_PTR(-ENOMEM);
180 lud = &vdv->vdv_cl.cd_lu_dev;
181 cl_device_init(&vdv->vdv_cl, t);
182 vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
183 vdv->vdv_cl.cd_ops = &vvp_cl_ops;
185 site = kzalloc(sizeof(*site), GFP_NOFS);
187 rc = cl_site_init(site, &vdv->vdv_cl);
189 rc = lu_site_init_finish(&site->cs_lu);
191 LASSERT(!lud->ld_site);
192 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;
212 vdv->vdv_next = lu2cl_dev(next);
214 LASSERT(d->ld_site && next->ld_type);
215 next->ld_site = d->ld_site;
216 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
217 next->ld_type->ldt_name,
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);
275 /*****************************************************************************
277 * mirror obd-devices into cl devices.
281 int cl_sb_init(struct super_block *sb)
283 struct ll_sb_info *sbi;
284 struct cl_device *cl;
290 env = cl_env_get(&refcheck);
292 cl = cl_type_setup(env, NULL, &vvp_device_type,
293 sbi->ll_dt_exp->exp_obd->obd_lu_dev);
295 cl2vvp_dev(cl)->vdv_sb = sb;
297 sbi->ll_site = cl2lu_dev(cl)->ld_site;
299 cl_env_put(env, &refcheck);
305 int cl_sb_fini(struct super_block *sb)
307 struct ll_sb_info *sbi;
309 struct cl_device *cld;
314 env = cl_env_get(&refcheck);
319 cl_stack_fini(env, cld);
323 cl_env_put(env, &refcheck);
326 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
327 result = PTR_ERR(env);
330 * If mount failed (sbi->ll_cl == NULL), and this there are no other
331 * mounts, stop device types manually (this usually happens
332 * automatically when last device is destroyed).
338 /****************************************************************************
340 * debugfs/lustre/llite/$MNT/dump_page_cache
342 ****************************************************************************/
345 * To represent contents of a page cache as a byte stream, following
346 * information if encoded in 64bit offset:
348 * - file hash bucket in lu_site::ls_hash[] 28bits
350 * - how far file is from bucket head 4bits
352 * - page index 32bits
354 * First two data identify a file in the cache uniquely.
357 #define PGC_OBJ_SHIFT (32 + 4)
358 #define PGC_DEPTH_SHIFT (32)
360 struct vvp_pgcache_id {
366 struct lu_object_header *vpi_obj;
369 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
371 CLASSERT(sizeof(pos) == sizeof(__u64));
373 id->vpi_index = pos & 0xffffffff;
374 id->vpi_depth = (pos >> PGC_DEPTH_SHIFT) & 0xf;
375 id->vpi_bucket = (unsigned long long)pos >> PGC_OBJ_SHIFT;
378 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
381 ((__u64)id->vpi_index) |
382 ((__u64)id->vpi_depth << PGC_DEPTH_SHIFT) |
383 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
386 static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
387 struct hlist_node *hnode, void *data)
389 struct vvp_pgcache_id *id = data;
390 struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
392 if (id->vpi_curdep-- > 0)
393 return 0; /* continue */
395 if (lu_object_is_dying(hdr))
398 cfs_hash_get(hs, hnode);
403 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
404 struct lu_device *dev,
405 struct vvp_pgcache_id *id)
407 LASSERT(lu_device_is_cl(dev));
409 id->vpi_depth &= 0xf;
411 id->vpi_curdep = id->vpi_depth;
413 cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
414 vvp_pgcache_obj_get, id);
416 struct lu_object *lu_obj;
418 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
420 lu_object_ref_add(lu_obj, "dump", current);
421 return lu2cl(lu_obj);
423 lu_object_put(env, lu_object_top(id->vpi_obj));
425 } else if (id->vpi_curdep > 0) {
431 static loff_t vvp_pgcache_find(const struct lu_env *env,
432 struct lu_device *dev, loff_t pos)
434 struct cl_object *clob;
435 struct lu_site *site;
436 struct vvp_pgcache_id id;
439 vvp_pgcache_id_unpack(pos, &id);
442 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
444 clob = vvp_pgcache_obj(env, dev, &id);
446 struct inode *inode = vvp_object_inode(clob);
450 nr = find_get_pages_contig(inode->i_mapping,
451 id.vpi_index, 1, &vmpage);
453 id.vpi_index = vmpage->index;
454 /* Cant support over 16T file */
455 nr = !(vmpage->index > 0xffffffff);
456 page_cache_release(vmpage);
459 lu_object_ref_del(&clob->co_lu, "dump", current);
460 cl_object_put(env, clob);
462 return vvp_pgcache_id_pack(&id);
464 /* to the next object. */
467 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
473 #define seq_page_flag(seq, page, flag, has_flags) do { \
474 if (test_bit(PG_##flag, &(page)->flags)) { \
475 seq_printf(seq, "%s"#flag, has_flags ? "|" : ""); \
480 static void vvp_pgcache_page_show(const struct lu_env *env,
481 struct seq_file *seq, struct cl_page *page)
483 struct vvp_page *vpg;
487 vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
488 vmpage = vpg->vpg_page;
489 seq_printf(seq, " %5i | %p %p %s %s %s %s | %p %lu/%u(%p) %lu %u [",
493 vpg->vpg_write_queued ? "wq" : "- ",
494 vpg->vpg_defer_uptodate ? "du" : "- ",
495 PageWriteback(vmpage) ? "wb" : "-",
496 vmpage, vmpage->mapping->host->i_ino,
497 vmpage->mapping->host->i_generation,
498 vmpage->mapping->host, vmpage->index,
501 seq_page_flag(seq, vmpage, locked, has_flags);
502 seq_page_flag(seq, vmpage, error, has_flags);
503 seq_page_flag(seq, vmpage, referenced, has_flags);
504 seq_page_flag(seq, vmpage, uptodate, has_flags);
505 seq_page_flag(seq, vmpage, dirty, has_flags);
506 seq_page_flag(seq, vmpage, writeback, has_flags);
507 seq_printf(seq, "%s]\n", has_flags ? "" : "-");
510 static int vvp_pgcache_show(struct seq_file *f, void *v)
513 struct ll_sb_info *sbi;
514 struct cl_object *clob;
516 struct vvp_pgcache_id id;
520 env = cl_env_get(&refcheck);
523 vvp_pgcache_id_unpack(pos, &id);
525 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
527 struct inode *inode = vvp_object_inode(clob);
528 struct cl_page *page = NULL;
531 result = find_get_pages_contig(inode->i_mapping,
536 page = cl_vmpage_page(vmpage, clob);
539 page_cache_release(vmpage);
542 seq_printf(f, "%8x@" DFID ": ", id.vpi_index,
543 PFID(lu_object_fid(&clob->co_lu)));
545 vvp_pgcache_page_show(env, f, page);
546 cl_page_put(env, page);
548 seq_puts(f, "missing\n");
549 lu_object_ref_del(&clob->co_lu, "dump", current);
550 cl_object_put(env, clob);
552 seq_printf(f, "%llx missing\n", pos);
553 cl_env_put(env, &refcheck);
556 result = PTR_ERR(env);
560 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
562 struct ll_sb_info *sbi;
568 env = cl_env_get(&refcheck);
571 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
572 pos = ERR_PTR(-EFBIG);
574 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
579 cl_env_put(env, &refcheck);
584 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
586 struct ll_sb_info *sbi;
590 env = cl_env_get(&refcheck);
593 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
596 cl_env_put(env, &refcheck);
601 static void vvp_pgcache_stop(struct seq_file *f, void *v)
606 static const struct seq_operations vvp_pgcache_ops = {
607 .start = vvp_pgcache_start,
608 .next = vvp_pgcache_next,
609 .stop = vvp_pgcache_stop,
610 .show = vvp_pgcache_show
613 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
615 struct seq_file *seq;
618 rc = seq_open(filp, &vvp_pgcache_ops);
622 seq = filp->private_data;
623 seq->private = inode->i_private;
628 const struct file_operations vvp_dump_pgcache_file_ops = {
629 .owner = THIS_MODULE,
630 .open = vvp_dump_pgcache_seq_open,
633 .release = seq_release,