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, 2013, 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_alloc(const struct lu_env *env,
139 struct lu_device_type *t,
140 struct lustre_cfg *cfg)
142 return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
145 static const struct lu_device_type_operations vvp_device_type_ops = {
146 .ldto_init = vvp_type_init,
147 .ldto_fini = vvp_type_fini,
149 .ldto_start = vvp_type_start,
150 .ldto_stop = vvp_type_stop,
152 .ldto_device_alloc = vvp_device_alloc,
153 .ldto_device_free = ccc_device_free,
154 .ldto_device_init = ccc_device_init,
155 .ldto_device_fini = ccc_device_fini
158 struct lu_device_type vvp_device_type = {
159 .ldt_tags = LU_DEVICE_CL,
160 .ldt_name = LUSTRE_VVP_NAME,
161 .ldt_ops = &vvp_device_type_ops,
162 .ldt_ctx_tags = LCT_CL_THREAD
166 * A mutex serializing calls to vvp_inode_fini() under extreme memory
167 * pressure, when environments cannot be allocated.
169 int vvp_global_init(void)
173 result = lu_kmem_init(vvp_caches);
175 result = ccc_global_init(&vvp_device_type);
177 lu_kmem_fini(vvp_caches);
182 void vvp_global_fini(void)
184 ccc_global_fini(&vvp_device_type);
185 lu_kmem_fini(vvp_caches);
189 /*****************************************************************************
191 * mirror obd-devices into cl devices.
195 int cl_sb_init(struct super_block *sb)
197 struct ll_sb_info *sbi;
198 struct cl_device *cl;
204 env = cl_env_get(&refcheck);
206 cl = cl_type_setup(env, NULL, &vvp_device_type,
207 sbi->ll_dt_exp->exp_obd->obd_lu_dev);
209 cl2ccc_dev(cl)->cdv_sb = sb;
211 sbi->ll_site = cl2lu_dev(cl)->ld_site;
213 cl_env_put(env, &refcheck);
219 int cl_sb_fini(struct super_block *sb)
221 struct ll_sb_info *sbi;
223 struct cl_device *cld;
229 env = cl_env_get(&refcheck);
234 cl_stack_fini(env, cld);
238 cl_env_put(env, &refcheck);
241 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
242 result = PTR_ERR(env);
248 /****************************************************************************
250 * /proc/fs/lustre/llite/$MNT/dump_page_cache
252 ****************************************************************************/
255 * To represent contents of a page cache as a byte stream, following
256 * information if encoded in 64bit offset:
258 * - file hash bucket in lu_site::ls_hash[] 28bits
260 * - how far file is from bucket head 4bits
262 * - page index 32bits
264 * First two data identify a file in the cache uniquely.
267 #define PGC_OBJ_SHIFT (32 + 4)
268 #define PGC_DEPTH_SHIFT (32)
270 struct vvp_pgcache_id {
276 struct lu_object_header *vpi_obj;
279 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
281 CLASSERT(sizeof(pos) == sizeof(__u64));
283 id->vpi_index = pos & 0xffffffff;
284 id->vpi_depth = (pos >> PGC_DEPTH_SHIFT) & 0xf;
285 id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
288 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
291 ((__u64)id->vpi_index) |
292 ((__u64)id->vpi_depth << PGC_DEPTH_SHIFT) |
293 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
296 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
297 struct hlist_node *hnode, void *data)
299 struct vvp_pgcache_id *id = data;
300 struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
302 if (id->vpi_curdep-- > 0)
303 return 0; /* continue */
305 if (lu_object_is_dying(hdr))
308 cfs_hash_get(hs, hnode);
313 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
314 struct lu_device *dev,
315 struct vvp_pgcache_id *id)
317 LASSERT(lu_device_is_cl(dev));
319 id->vpi_depth &= 0xf;
321 id->vpi_curdep = id->vpi_depth;
323 cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
324 vvp_pgcache_obj_get, id);
325 if (id->vpi_obj != NULL) {
326 struct lu_object *lu_obj;
328 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
329 if (lu_obj != NULL) {
330 lu_object_ref_add(lu_obj, "dump", current);
331 return lu2cl(lu_obj);
333 lu_object_put(env, lu_object_top(id->vpi_obj));
335 } else if (id->vpi_curdep > 0) {
341 static loff_t vvp_pgcache_find(const struct lu_env *env,
342 struct lu_device *dev, loff_t pos)
344 struct cl_object *clob;
345 struct lu_site *site;
346 struct vvp_pgcache_id id;
349 vvp_pgcache_id_unpack(pos, &id);
352 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
354 clob = vvp_pgcache_obj(env, dev, &id);
356 struct inode *inode = ccc_object_inode(clob);
360 nr = find_get_pages_contig(inode->i_mapping,
361 id.vpi_index, 1, &vmpage);
363 id.vpi_index = vmpage->index;
364 /* Cant support over 16T file */
365 nr = !(vmpage->index > 0xffffffff);
366 page_cache_release(vmpage);
369 lu_object_ref_del(&clob->co_lu, "dump", current);
370 cl_object_put(env, clob);
372 return vvp_pgcache_id_pack(&id);
374 /* to the next object. */
377 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
383 #define seq_page_flag(seq, page, flag, has_flags) do { \
384 if (test_bit(PG_##flag, &(page)->flags)) { \
385 seq_printf(seq, "%s"#flag, has_flags ? "|" : ""); \
390 static void vvp_pgcache_page_show(const struct lu_env *env,
391 struct seq_file *seq, struct cl_page *page)
393 struct ccc_page *cpg;
397 cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
398 vmpage = cpg->cpg_page;
399 seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
403 cpg->cpg_write_queued ? "wq" : "- ",
404 cpg->cpg_defer_uptodate ? "du" : "- ",
405 PageWriteback(vmpage) ? "wb" : "-",
407 PFID(ll_inode2fid(vmpage->mapping->host)),
408 vmpage->mapping->host, vmpage->index,
411 seq_page_flag(seq, vmpage, locked, has_flags);
412 seq_page_flag(seq, vmpage, error, has_flags);
413 seq_page_flag(seq, vmpage, referenced, has_flags);
414 seq_page_flag(seq, vmpage, uptodate, has_flags);
415 seq_page_flag(seq, vmpage, dirty, has_flags);
416 seq_page_flag(seq, vmpage, writeback, has_flags);
417 seq_printf(seq, "%s]\n", has_flags ? "" : "-");
420 static int vvp_pgcache_show(struct seq_file *f, void *v)
423 struct ll_sb_info *sbi;
424 struct cl_object *clob;
426 struct vvp_pgcache_id id;
430 env = cl_env_get(&refcheck);
433 vvp_pgcache_id_unpack(pos, &id);
435 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
437 struct inode *inode = ccc_object_inode(clob);
438 struct cl_page *page = NULL;
441 result = find_get_pages_contig(inode->i_mapping,
442 id.vpi_index, 1, &vmpage);
445 page = cl_vmpage_page(vmpage, clob);
448 page_cache_release(vmpage);
451 seq_printf(f, "%8x@"DFID": ", id.vpi_index,
452 PFID(lu_object_fid(&clob->co_lu)));
454 vvp_pgcache_page_show(env, f, page);
455 cl_page_put(env, page);
457 seq_puts(f, "missing\n");
458 lu_object_ref_del(&clob->co_lu, "dump", current);
459 cl_object_put(env, clob);
461 seq_printf(f, "%llx missing\n", pos);
462 cl_env_put(env, &refcheck);
465 result = PTR_ERR(env);
469 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
471 struct ll_sb_info *sbi;
477 env = cl_env_get(&refcheck);
480 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
481 pos = ERR_PTR(-EFBIG);
483 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
488 cl_env_put(env, &refcheck);
493 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
495 struct ll_sb_info *sbi;
499 env = cl_env_get(&refcheck);
502 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
505 cl_env_put(env, &refcheck);
510 static void vvp_pgcache_stop(struct seq_file *f, void *v)
515 static struct seq_operations vvp_pgcache_ops = {
516 .start = vvp_pgcache_start,
517 .next = vvp_pgcache_next,
518 .stop = vvp_pgcache_stop,
519 .show = vvp_pgcache_show
522 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
524 struct ll_sb_info *sbi = PDE_DATA(inode);
525 struct seq_file *seq;
528 result = seq_open(filp, &vvp_pgcache_ops);
530 seq = filp->private_data;
536 const struct file_operations vvp_dump_pgcache_file_ops = {
537 .owner = THIS_MODULE,
538 .open = vvp_dump_pgcache_seq_open,
541 .release = seq_release,