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
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 static struct kmem_cache *vvp_session_kmem;
59 static struct kmem_cache *vvp_thread_kmem;
61 static struct lu_kmem_descr vvp_caches[] = {
63 .ckd_cache = &ll_thread_kmem,
64 .ckd_name = "ll_thread_kmem",
65 .ckd_size = sizeof(struct ll_thread_info),
68 .ckd_cache = &vvp_lock_kmem,
69 .ckd_name = "vvp_lock_kmem",
70 .ckd_size = sizeof(struct vvp_lock),
73 .ckd_cache = &vvp_object_kmem,
74 .ckd_name = "vvp_object_kmem",
75 .ckd_size = sizeof(struct vvp_object),
78 .ckd_cache = &vvp_session_kmem,
79 .ckd_name = "vvp_session_kmem",
80 .ckd_size = sizeof(struct vvp_session)
83 .ckd_cache = &vvp_thread_kmem,
84 .ckd_name = "vvp_thread_kmem",
85 .ckd_size = sizeof(struct vvp_thread_info),
92 static void *ll_thread_key_init(const struct lu_context *ctx,
93 struct lu_context_key *key)
95 struct vvp_thread_info *info;
97 info = kmem_cache_zalloc(ll_thread_kmem, GFP_NOFS);
99 info = ERR_PTR(-ENOMEM);
103 static void ll_thread_key_fini(const struct lu_context *ctx,
104 struct lu_context_key *key, void *data)
106 struct vvp_thread_info *info = data;
108 kmem_cache_free(ll_thread_kmem, info);
111 struct lu_context_key ll_thread_key = {
112 .lct_tags = LCT_CL_THREAD,
113 .lct_init = ll_thread_key_init,
114 .lct_fini = ll_thread_key_fini
117 static void *vvp_session_key_init(const struct lu_context *ctx,
118 struct lu_context_key *key)
120 struct vvp_session *session;
122 session = kmem_cache_zalloc(vvp_session_kmem, GFP_NOFS);
124 session = ERR_PTR(-ENOMEM);
128 static void vvp_session_key_fini(const struct lu_context *ctx,
129 struct lu_context_key *key, void *data)
131 struct vvp_session *session = data;
133 kmem_cache_free(vvp_session_kmem, session);
136 struct lu_context_key vvp_session_key = {
137 .lct_tags = LCT_SESSION,
138 .lct_init = vvp_session_key_init,
139 .lct_fini = vvp_session_key_fini
142 static void *vvp_thread_key_init(const struct lu_context *ctx,
143 struct lu_context_key *key)
145 struct vvp_thread_info *vti;
147 vti = kmem_cache_zalloc(vvp_thread_kmem, GFP_NOFS);
149 vti = ERR_PTR(-ENOMEM);
153 static void vvp_thread_key_fini(const struct lu_context *ctx,
154 struct lu_context_key *key, void *data)
156 struct vvp_thread_info *vti = data;
158 kmem_cache_free(vvp_thread_kmem, vti);
161 struct lu_context_key vvp_thread_key = {
162 .lct_tags = LCT_CL_THREAD,
163 .lct_init = vvp_thread_key_init,
164 .lct_fini = vvp_thread_key_fini
167 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
168 LU_TYPE_INIT_FINI(vvp, &vvp_thread_key, &ll_thread_key, &vvp_session_key);
170 static const struct lu_device_operations vvp_lu_ops = {
171 .ldo_object_alloc = vvp_object_alloc
174 static struct lu_device *vvp_device_free(const struct lu_env *env,
177 struct vvp_device *vdv = lu2vvp_dev(d);
178 struct cl_site *site = lu2cl_site(d->ld_site);
179 struct lu_device *next = cl2lu_dev(vdv->vdv_next);
185 cl_device_fini(lu2cl_dev(d));
190 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
191 struct lu_device_type *t,
192 struct lustre_cfg *cfg)
194 struct vvp_device *vdv;
195 struct lu_device *lud;
196 struct cl_site *site;
199 vdv = kzalloc(sizeof(*vdv), GFP_NOFS);
201 return ERR_PTR(-ENOMEM);
203 lud = &vdv->vdv_cl.cd_lu_dev;
204 cl_device_init(&vdv->vdv_cl, t);
205 vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
207 site = kzalloc(sizeof(*site), GFP_NOFS);
209 rc = cl_site_init(site, &vdv->vdv_cl);
211 rc = lu_site_init_finish(&site->cs_lu);
213 LASSERT(!lud->ld_site);
214 CERROR("Cannot init lu_site, rc %d.\n", rc);
221 vvp_device_free(env, lud);
227 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
228 const char *name, struct lu_device *next)
230 struct vvp_device *vdv;
234 vdv->vdv_next = lu2cl_dev(next);
236 LASSERT(d->ld_site && next->ld_type);
237 next->ld_site = d->ld_site;
238 rc = next->ld_type->ldt_ops->ldto_device_init(env, next,
239 next->ld_type->ldt_name,
243 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
248 static struct lu_device *vvp_device_fini(const struct lu_env *env,
251 return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
254 static const struct lu_device_type_operations vvp_device_type_ops = {
255 .ldto_init = vvp_type_init,
256 .ldto_fini = vvp_type_fini,
258 .ldto_start = vvp_type_start,
259 .ldto_stop = vvp_type_stop,
261 .ldto_device_alloc = vvp_device_alloc,
262 .ldto_device_free = vvp_device_free,
263 .ldto_device_init = vvp_device_init,
264 .ldto_device_fini = vvp_device_fini,
267 struct lu_device_type vvp_device_type = {
268 .ldt_tags = LU_DEVICE_CL,
269 .ldt_name = LUSTRE_VVP_NAME,
270 .ldt_ops = &vvp_device_type_ops,
271 .ldt_ctx_tags = LCT_CL_THREAD
275 * A mutex serializing calls to vvp_inode_fini() under extreme memory
276 * pressure, when environments cannot be allocated.
278 int vvp_global_init(void)
282 rc = lu_kmem_init(vvp_caches);
286 rc = lu_device_type_init(&vvp_device_type);
293 lu_kmem_fini(vvp_caches);
298 void vvp_global_fini(void)
300 lu_device_type_fini(&vvp_device_type);
301 lu_kmem_fini(vvp_caches);
304 /*****************************************************************************
306 * mirror obd-devices into cl devices.
310 int cl_sb_init(struct super_block *sb)
312 struct ll_sb_info *sbi;
313 struct cl_device *cl;
319 env = cl_env_get(&refcheck);
321 cl = cl_type_setup(env, NULL, &vvp_device_type,
322 sbi->ll_dt_exp->exp_obd->obd_lu_dev);
325 sbi->ll_site = cl2lu_dev(cl)->ld_site;
327 cl_env_put(env, &refcheck);
334 int cl_sb_fini(struct super_block *sb)
336 struct ll_sb_info *sbi;
338 struct cl_device *cld;
343 env = cl_env_get(&refcheck);
348 cl_stack_fini(env, cld);
352 cl_env_put(env, &refcheck);
355 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
356 result = PTR_ERR(env);
361 /****************************************************************************
363 * debugfs/lustre/llite/$MNT/dump_page_cache
365 ****************************************************************************/
368 * To represent contents of a page cache as a byte stream, following
369 * information if encoded in 64bit offset:
371 * - file hash bucket in lu_site::ls_hash[] 28bits
373 * - how far file is from bucket head 4bits
375 * - page index 32bits
377 * First two data identify a file in the cache uniquely.
380 #define PGC_OBJ_SHIFT (32 + 4)
381 #define PGC_DEPTH_SHIFT (32)
383 struct vvp_pgcache_id {
384 unsigned int vpi_bucket;
385 unsigned int vpi_depth;
388 unsigned int vpi_curdep;
389 struct lu_object_header *vpi_obj;
392 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
394 BUILD_BUG_ON(sizeof(pos) != sizeof(__u64));
396 id->vpi_index = pos & 0xffffffff;
397 id->vpi_depth = (pos >> PGC_DEPTH_SHIFT) & 0xf;
398 id->vpi_bucket = (unsigned long long)pos >> PGC_OBJ_SHIFT;
401 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
404 ((__u64)id->vpi_index) |
405 ((__u64)id->vpi_depth << PGC_DEPTH_SHIFT) |
406 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
409 static int vvp_pgcache_obj_get(struct cfs_hash *hs, struct cfs_hash_bd *bd,
410 struct hlist_node *hnode, void *data)
412 struct vvp_pgcache_id *id = data;
413 struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
415 if (id->vpi_curdep-- > 0)
416 return 0; /* continue */
418 if (lu_object_is_dying(hdr))
421 cfs_hash_get(hs, hnode);
426 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
427 struct lu_device *dev,
428 struct vvp_pgcache_id *id)
430 LASSERT(lu_device_is_cl(dev));
432 id->vpi_depth &= 0xf;
434 id->vpi_curdep = id->vpi_depth;
436 cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
437 vvp_pgcache_obj_get, id);
439 struct lu_object *lu_obj;
441 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
443 lu_object_ref_add(lu_obj, "dump", current);
444 return lu2cl(lu_obj);
446 lu_object_put(env, lu_object_top(id->vpi_obj));
448 } else if (id->vpi_curdep > 0) {
454 static loff_t vvp_pgcache_find(const struct lu_env *env,
455 struct lu_device *dev, loff_t pos)
457 struct cl_object *clob;
458 struct lu_site *site;
459 struct vvp_pgcache_id id;
462 vvp_pgcache_id_unpack(pos, &id);
465 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
467 clob = vvp_pgcache_obj(env, dev, &id);
469 struct inode *inode = vvp_object_inode(clob);
473 nr = find_get_pages_contig(inode->i_mapping,
474 id.vpi_index, 1, &vmpage);
476 id.vpi_index = vmpage->index;
477 /* Cant support over 16T file */
478 nr = !(vmpage->index > 0xffffffff);
482 lu_object_ref_del(&clob->co_lu, "dump", current);
483 cl_object_put(env, clob);
485 return vvp_pgcache_id_pack(&id);
487 /* to the next object. */
490 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
496 #define seq_page_flag(seq, page, flag, has_flags) do { \
497 if (test_bit(PG_##flag, &(page)->flags)) { \
498 seq_printf(seq, "%s"#flag, has_flags ? "|" : ""); \
503 static void vvp_pgcache_page_show(const struct lu_env *env,
504 struct seq_file *seq, struct cl_page *page)
506 struct vvp_page *vpg;
510 vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
511 vmpage = vpg->vpg_page;
512 seq_printf(seq, " %5i | %p %p %s %s %s | %p " DFID "(%p) %lu %u [",
516 vpg->vpg_defer_uptodate ? "du" : "- ",
517 PageWriteback(vmpage) ? "wb" : "-",
518 vmpage, PFID(ll_inode2fid(vmpage->mapping->host)),
519 vmpage->mapping->host, vmpage->index,
522 seq_page_flag(seq, vmpage, locked, has_flags);
523 seq_page_flag(seq, vmpage, error, has_flags);
524 seq_page_flag(seq, vmpage, referenced, has_flags);
525 seq_page_flag(seq, vmpage, uptodate, has_flags);
526 seq_page_flag(seq, vmpage, dirty, has_flags);
527 seq_page_flag(seq, vmpage, writeback, has_flags);
528 seq_printf(seq, "%s]\n", has_flags ? "" : "-");
531 static int vvp_pgcache_show(struct seq_file *f, void *v)
534 struct ll_sb_info *sbi;
535 struct cl_object *clob;
537 struct vvp_pgcache_id id;
541 env = cl_env_get(&refcheck);
544 vvp_pgcache_id_unpack(pos, &id);
546 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
548 struct inode *inode = vvp_object_inode(clob);
549 struct cl_page *page = NULL;
552 result = find_get_pages_contig(inode->i_mapping,
557 page = cl_vmpage_page(vmpage, clob);
562 seq_printf(f, "%8x@" DFID ": ", id.vpi_index,
563 PFID(lu_object_fid(&clob->co_lu)));
565 vvp_pgcache_page_show(env, f, page);
566 cl_page_put(env, page);
568 seq_puts(f, "missing\n");
570 lu_object_ref_del(&clob->co_lu, "dump", current);
571 cl_object_put(env, clob);
573 seq_printf(f, "%llx missing\n", pos);
575 cl_env_put(env, &refcheck);
578 result = PTR_ERR(env);
583 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
585 struct ll_sb_info *sbi;
591 env = cl_env_get(&refcheck);
594 if (sbi->ll_site->ls_obj_hash->hs_cur_bits >
595 64 - PGC_OBJ_SHIFT) {
596 pos = ERR_PTR(-EFBIG);
598 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
603 cl_env_put(env, &refcheck);
608 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
610 struct ll_sb_info *sbi;
614 env = cl_env_get(&refcheck);
617 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
620 cl_env_put(env, &refcheck);
625 static void vvp_pgcache_stop(struct seq_file *f, void *v)
630 static const struct seq_operations vvp_pgcache_ops = {
631 .start = vvp_pgcache_start,
632 .next = vvp_pgcache_next,
633 .stop = vvp_pgcache_stop,
634 .show = vvp_pgcache_show
637 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
639 struct seq_file *seq;
642 rc = seq_open(filp, &vvp_pgcache_ops);
646 seq = filp->private_data;
647 seq->private = inode->i_private;
652 const struct file_operations vvp_dump_pgcache_file_ops = {
653 .owner = THIS_MODULE,
654 .open = vvp_dump_pgcache_seq_open,
657 .release = seq_release,