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 static struct kmem_cache *vvp_thread_kmem;
61 static struct kmem_cache *vvp_session_kmem;
62 static struct lu_kmem_descr vvp_caches[] = {
64 .ckd_cache = &vvp_thread_kmem,
65 .ckd_name = "vvp_thread_kmem",
66 .ckd_size = sizeof (struct vvp_thread_info),
69 .ckd_cache = &vvp_session_kmem,
70 .ckd_name = "vvp_session_kmem",
71 .ckd_size = sizeof (struct vvp_session)
78 static void *vvp_key_init(const struct lu_context *ctx,
79 struct lu_context_key *key)
81 struct vvp_thread_info *info;
83 OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, GFP_NOFS);
85 info = ERR_PTR(-ENOMEM);
89 static void vvp_key_fini(const struct lu_context *ctx,
90 struct lu_context_key *key, void *data)
92 struct vvp_thread_info *info = data;
93 OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
96 static void *vvp_session_key_init(const struct lu_context *ctx,
97 struct lu_context_key *key)
99 struct vvp_session *session;
101 OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
103 session = ERR_PTR(-ENOMEM);
107 static void vvp_session_key_fini(const struct lu_context *ctx,
108 struct lu_context_key *key, void *data)
110 struct vvp_session *session = data;
111 OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
115 struct lu_context_key vvp_key = {
116 .lct_tags = LCT_CL_THREAD,
117 .lct_init = vvp_key_init,
118 .lct_fini = vvp_key_fini
121 struct lu_context_key vvp_session_key = {
122 .lct_tags = LCT_SESSION,
123 .lct_init = vvp_session_key_init,
124 .lct_fini = vvp_session_key_fini
127 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
128 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
130 static const struct lu_device_operations vvp_lu_ops = {
131 .ldo_object_alloc = vvp_object_alloc
134 static const struct cl_device_operations vvp_cl_ops = {
135 .cdo_req_init = ccc_req_init
138 static struct lu_device *vvp_device_free(const struct lu_env *env,
141 struct vvp_device *vdv = lu2vvp_dev(d);
142 struct cl_site *site = lu2cl_site(d->ld_site);
143 struct lu_device *next = cl2lu_dev(vdv->vdv_next);
145 if (d->ld_site != NULL) {
150 cl_device_fini(lu2cl_dev(d));
155 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
156 struct lu_device_type *t,
157 struct lustre_cfg *cfg)
159 struct vvp_device *vdv;
160 struct lu_device *lud;
161 struct cl_site *site;
167 RETURN(ERR_PTR(-ENOMEM));
169 lud = &vdv->vdv_cl.cd_lu_dev;
170 cl_device_init(&vdv->vdv_cl, t);
171 vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
172 vdv->vdv_cl.cd_ops = &vvp_cl_ops;
176 rc = cl_site_init(site, &vdv->vdv_cl);
178 rc = lu_site_init_finish(&site->cs_lu);
180 LASSERT(lud->ld_site == NULL);
181 CERROR("Cannot init lu_site, rc %d.\n", rc);
187 vvp_device_free(env, lud);
193 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
194 const char *name, struct lu_device *next)
196 struct vvp_device *vdv;
201 vdv->vdv_next = lu2cl_dev(next);
203 LASSERT(d->ld_site != NULL && next->ld_type != NULL);
204 next->ld_site = d->ld_site;
205 rc = next->ld_type->ldt_ops->ldto_device_init(
206 env, next, next->ld_type->ldt_name, NULL);
209 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
214 static struct lu_device *vvp_device_fini(const struct lu_env *env,
217 return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
220 static const struct lu_device_type_operations vvp_device_type_ops = {
221 .ldto_init = vvp_type_init,
222 .ldto_fini = vvp_type_fini,
224 .ldto_start = vvp_type_start,
225 .ldto_stop = vvp_type_stop,
227 .ldto_device_alloc = vvp_device_alloc,
228 .ldto_device_free = vvp_device_free,
229 .ldto_device_init = vvp_device_init,
230 .ldto_device_fini = vvp_device_fini,
233 struct lu_device_type vvp_device_type = {
234 .ldt_tags = LU_DEVICE_CL,
235 .ldt_name = LUSTRE_VVP_NAME,
236 .ldt_ops = &vvp_device_type_ops,
237 .ldt_ctx_tags = LCT_CL_THREAD
241 * A mutex serializing calls to vvp_inode_fini() under extreme memory
242 * pressure, when environments cannot be allocated.
244 int vvp_global_init(void)
248 result = lu_kmem_init(vvp_caches);
250 result = ccc_global_init(&vvp_device_type);
252 lu_kmem_fini(vvp_caches);
257 void vvp_global_fini(void)
259 ccc_global_fini(&vvp_device_type);
260 lu_kmem_fini(vvp_caches);
264 /*****************************************************************************
266 * mirror obd-devices into cl devices.
270 int cl_sb_init(struct super_block *sb)
272 struct ll_sb_info *sbi;
273 struct cl_device *cl;
279 env = cl_env_get(&refcheck);
281 cl = cl_type_setup(env, NULL, &vvp_device_type,
282 sbi->ll_dt_exp->exp_obd->obd_lu_dev);
284 cl2vvp_dev(cl)->vdv_sb = sb;
286 sbi->ll_site = cl2lu_dev(cl)->ld_site;
288 cl_env_put(env, &refcheck);
294 int cl_sb_fini(struct super_block *sb)
296 struct ll_sb_info *sbi;
298 struct cl_device *cld;
304 env = cl_env_get(&refcheck);
309 cl_stack_fini(env, cld);
313 cl_env_put(env, &refcheck);
316 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
317 result = PTR_ERR(env);
323 /****************************************************************************
325 * /proc/fs/lustre/llite/$MNT/dump_page_cache
327 ****************************************************************************/
330 * To represent contents of a page cache as a byte stream, following
331 * information if encoded in 64bit offset:
333 * - file hash bucket in lu_site::ls_hash[] 28bits
335 * - how far file is from bucket head 4bits
337 * - page index 32bits
339 * First two data identify a file in the cache uniquely.
342 #define PGC_OBJ_SHIFT (32 + 4)
343 #define PGC_DEPTH_SHIFT (32)
345 struct vvp_pgcache_id {
351 struct lu_object_header *vpi_obj;
354 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
356 CLASSERT(sizeof(pos) == sizeof(__u64));
358 id->vpi_index = pos & 0xffffffff;
359 id->vpi_depth = (pos >> PGC_DEPTH_SHIFT) & 0xf;
360 id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
363 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
366 ((__u64)id->vpi_index) |
367 ((__u64)id->vpi_depth << PGC_DEPTH_SHIFT) |
368 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
371 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
372 struct hlist_node *hnode, void *data)
374 struct vvp_pgcache_id *id = data;
375 struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
377 if (id->vpi_curdep-- > 0)
378 return 0; /* continue */
380 if (lu_object_is_dying(hdr))
383 cfs_hash_get(hs, hnode);
388 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
389 struct lu_device *dev,
390 struct vvp_pgcache_id *id)
392 LASSERT(lu_device_is_cl(dev));
394 id->vpi_depth &= 0xf;
396 id->vpi_curdep = id->vpi_depth;
398 cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
399 vvp_pgcache_obj_get, id);
400 if (id->vpi_obj != NULL) {
401 struct lu_object *lu_obj;
403 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
404 if (lu_obj != NULL) {
405 lu_object_ref_add(lu_obj, "dump", current);
406 return lu2cl(lu_obj);
408 lu_object_put(env, lu_object_top(id->vpi_obj));
410 } else if (id->vpi_curdep > 0) {
416 static loff_t vvp_pgcache_find(const struct lu_env *env,
417 struct lu_device *dev, loff_t pos)
419 struct cl_object *clob;
420 struct lu_site *site;
421 struct vvp_pgcache_id id;
424 vvp_pgcache_id_unpack(pos, &id);
427 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
429 clob = vvp_pgcache_obj(env, dev, &id);
431 struct inode *inode = ccc_object_inode(clob);
435 nr = find_get_pages_contig(inode->i_mapping,
436 id.vpi_index, 1, &vmpage);
438 id.vpi_index = vmpage->index;
439 /* Cant support over 16T file */
440 nr = !(vmpage->index > 0xffffffff);
441 page_cache_release(vmpage);
444 lu_object_ref_del(&clob->co_lu, "dump", current);
445 cl_object_put(env, clob);
447 return vvp_pgcache_id_pack(&id);
449 /* to the next object. */
452 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
458 #define seq_page_flag(seq, page, flag, has_flags) do { \
459 if (test_bit(PG_##flag, &(page)->flags)) { \
460 seq_printf(seq, "%s"#flag, has_flags ? "|" : ""); \
465 static void vvp_pgcache_page_show(const struct lu_env *env,
466 struct seq_file *seq, struct cl_page *page)
468 struct ccc_page *cpg;
472 cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
473 vmpage = cpg->cpg_page;
474 seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
478 cpg->cpg_write_queued ? "wq" : "- ",
479 cpg->cpg_defer_uptodate ? "du" : "- ",
480 PageWriteback(vmpage) ? "wb" : "-",
482 PFID(ll_inode2fid(vmpage->mapping->host)),
483 vmpage->mapping->host, vmpage->index,
486 seq_page_flag(seq, vmpage, locked, has_flags);
487 seq_page_flag(seq, vmpage, error, has_flags);
488 seq_page_flag(seq, vmpage, referenced, has_flags);
489 seq_page_flag(seq, vmpage, uptodate, has_flags);
490 seq_page_flag(seq, vmpage, dirty, has_flags);
491 seq_page_flag(seq, vmpage, writeback, has_flags);
492 seq_printf(seq, "%s]\n", has_flags ? "" : "-");
495 static int vvp_pgcache_show(struct seq_file *f, void *v)
498 struct ll_sb_info *sbi;
499 struct cl_object *clob;
501 struct vvp_pgcache_id id;
505 env = cl_env_get(&refcheck);
508 vvp_pgcache_id_unpack(pos, &id);
510 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
512 struct inode *inode = ccc_object_inode(clob);
513 struct cl_page *page = NULL;
516 result = find_get_pages_contig(inode->i_mapping,
517 id.vpi_index, 1, &vmpage);
520 page = cl_vmpage_page(vmpage, clob);
523 page_cache_release(vmpage);
526 seq_printf(f, "%8x@"DFID": ", id.vpi_index,
527 PFID(lu_object_fid(&clob->co_lu)));
529 vvp_pgcache_page_show(env, f, page);
530 cl_page_put(env, page);
532 seq_puts(f, "missing\n");
533 lu_object_ref_del(&clob->co_lu, "dump", current);
534 cl_object_put(env, clob);
536 seq_printf(f, "%llx missing\n", pos);
537 cl_env_put(env, &refcheck);
540 result = PTR_ERR(env);
544 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
546 struct ll_sb_info *sbi;
552 env = cl_env_get(&refcheck);
555 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
556 pos = ERR_PTR(-EFBIG);
558 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
563 cl_env_put(env, &refcheck);
568 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
570 struct ll_sb_info *sbi;
574 env = cl_env_get(&refcheck);
577 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
580 cl_env_put(env, &refcheck);
585 static void vvp_pgcache_stop(struct seq_file *f, void *v)
590 static struct seq_operations vvp_pgcache_ops = {
591 .start = vvp_pgcache_start,
592 .next = vvp_pgcache_next,
593 .stop = vvp_pgcache_stop,
594 .show = vvp_pgcache_show
597 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
599 struct ll_sb_info *sbi = PDE_DATA(inode);
600 struct seq_file *seq;
603 result = seq_open(filp, &vvp_pgcache_ops);
605 seq = filp->private_data;
611 const struct file_operations vvp_dump_pgcache_file_ops = {
612 .owner = THIS_MODULE,
613 .open = vvp_dump_pgcache_seq_open,
616 .release = seq_release,