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.
31 * This file is part of Lustre, http://www.lustre.org/
32 * Lustre is a trademark of Sun Microsystems, Inc.
34 * cl_device and cl_device_type implementation for VVP layer.
36 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 #define DEBUG_SUBSYSTEM S_LLITE
42 # error This file is kernel only.
46 #include <lustre_lite.h>
48 #include "vvp_internal.h"
50 /*****************************************************************************
52 * Vvp device and device type functions.
57 * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
58 * "llite_" (var. "ll_") prefix.
61 cfs_mem_cache_t *vvp_page_kmem;
62 cfs_mem_cache_t *vvp_thread_kmem;
63 static cfs_mem_cache_t *vvp_session_kmem;
64 static struct lu_kmem_descr vvp_caches[] = {
66 .ckd_cache = &vvp_page_kmem,
67 .ckd_name = "vvp_page_kmem",
68 .ckd_size = sizeof (struct ccc_page)
71 .ckd_cache = &vvp_thread_kmem,
72 .ckd_name = "vvp_thread_kmem",
73 .ckd_size = sizeof (struct vvp_thread_info),
76 .ckd_cache = &vvp_session_kmem,
77 .ckd_name = "vvp_session_kmem",
78 .ckd_size = sizeof (struct vvp_session)
85 static void *vvp_key_init(const struct lu_context *ctx,
86 struct lu_context_key *key)
88 struct vvp_thread_info *info;
90 OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, CFS_ALLOC_IO);
92 info = ERR_PTR(-ENOMEM);
96 static void vvp_key_fini(const struct lu_context *ctx,
97 struct lu_context_key *key, void *data)
99 struct vvp_thread_info *info = data;
100 OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
103 static void *vvp_session_key_init(const struct lu_context *ctx,
104 struct lu_context_key *key)
106 struct vvp_session *session;
108 OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, CFS_ALLOC_IO);
110 session = ERR_PTR(-ENOMEM);
114 static void vvp_session_key_fini(const struct lu_context *ctx,
115 struct lu_context_key *key, void *data)
117 struct vvp_session *session = data;
118 OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
122 struct lu_context_key vvp_key = {
123 .lct_tags = LCT_CL_THREAD,
124 .lct_init = vvp_key_init,
125 .lct_fini = vvp_key_fini
128 struct lu_context_key vvp_session_key = {
129 .lct_tags = LCT_SESSION,
130 .lct_init = vvp_session_key_init,
131 .lct_fini = vvp_session_key_fini
134 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
135 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
137 static const struct lu_device_operations vvp_lu_ops = {
138 .ldo_object_alloc = vvp_object_alloc
141 static const struct cl_device_operations vvp_cl_ops = {
142 .cdo_req_init = ccc_req_init
145 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
146 struct lu_device_type *t,
147 struct lustre_cfg *cfg)
149 return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
152 static const struct lu_device_type_operations vvp_device_type_ops = {
153 .ldto_init = vvp_type_init,
154 .ldto_fini = vvp_type_fini,
156 .ldto_start = vvp_type_start,
157 .ldto_stop = vvp_type_stop,
159 .ldto_device_alloc = vvp_device_alloc,
160 .ldto_device_free = ccc_device_free,
161 .ldto_device_init = ccc_device_init,
162 .ldto_device_fini = ccc_device_fini
165 struct lu_device_type vvp_device_type = {
166 .ldt_tags = LU_DEVICE_CL,
167 .ldt_name = LUSTRE_VVP_NAME,
168 .ldt_ops = &vvp_device_type_ops,
169 .ldt_ctx_tags = LCT_CL_THREAD
173 * A mutex serializing calls to vvp_inode_fini() under extreme memory
174 * pressure, when environments cannot be allocated.
176 int vvp_global_init(void)
180 result = lu_kmem_init(vvp_caches);
182 result = ccc_global_init(&vvp_device_type);
184 lu_kmem_fini(vvp_caches);
189 void vvp_global_fini(void)
191 ccc_global_fini(&vvp_device_type);
192 lu_kmem_fini(vvp_caches);
196 /*****************************************************************************
198 * mirror obd-devices into cl devices.
202 int cl_sb_init(struct super_block *sb)
204 struct ll_sb_info *sbi;
205 struct cl_device *cl;
211 env = cl_env_get(&refcheck);
213 cl = cl_type_setup(env, NULL, &vvp_device_type,
214 sbi->ll_dt_exp->exp_obd->obd_lu_dev);
216 cl2ccc_dev(cl)->cdv_sb = sb;
218 sbi->ll_site = cl2lu_dev(cl)->ld_site;
220 cl_env_put(env, &refcheck);
226 int cl_sb_fini(struct super_block *sb)
228 struct ll_sb_info *sbi;
230 struct cl_device *cld;
236 env = cl_env_get(&refcheck);
241 cl_stack_fini(env, cld);
245 cl_env_put(env, &refcheck);
248 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
249 result = PTR_ERR(env);
252 * If mount failed (sbi->ll_cl == NULL), and this there are no other
253 * mounts, stop device types manually (this usually happens
254 * automatically when last device is destroyed).
260 /****************************************************************************
262 * /proc/fs/lustre/llite/$MNT/dump_page_cache
264 ****************************************************************************/
267 * To represent contents of a page cache as a byte stream, following
268 * information if encoded in 64bit offset:
270 * - file hash bucket in lu_site::ls_hash[] 28bits
272 * - how far file is from bucket head 4bits
274 * - page index 32bits
276 * First two data identify a file in the cache uniquely.
279 #define PGC_OBJ_SHIFT (32 + 4)
280 #define PGC_DEPTH_SHIFT (32)
282 struct vvp_pgcache_id {
288 struct lu_object_header *vpi_obj;
291 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
293 CLASSERT(sizeof(pos) == sizeof(__u64));
295 id->vpi_index = pos & 0xffffffff;
296 id->vpi_depth = (pos >> PGC_DEPTH_SHIFT) & 0xf;
297 id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
300 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
303 ((__u64)id->vpi_index) |
304 ((__u64)id->vpi_depth << PGC_DEPTH_SHIFT) |
305 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
308 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
309 cfs_hlist_node_t *hnode, void *data)
311 struct vvp_pgcache_id *id = data;
312 struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
314 if (id->vpi_curdep-- > 0)
315 return 0; /* continue */
317 if (lu_object_is_dying(hdr))
320 cfs_hash_get(hs, hnode);
325 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
326 struct lu_device *dev,
327 struct vvp_pgcache_id *id)
329 LASSERT(lu_device_is_cl(dev));
331 id->vpi_depth &= 0xf;
333 id->vpi_curdep = id->vpi_depth;
335 cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
336 vvp_pgcache_obj_get, id);
337 if (id->vpi_obj != NULL) {
338 struct lu_object *lu_obj;
340 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
341 if (lu_obj != NULL) {
342 lu_object_ref_add(lu_obj, "dump", cfs_current());
343 return lu2cl(lu_obj);
345 lu_object_put(env, lu_object_top(id->vpi_obj));
347 } else if (id->vpi_curdep > 0) {
353 static loff_t vvp_pgcache_find(const struct lu_env *env,
354 struct lu_device *dev, loff_t pos)
356 struct cl_object *clob;
357 struct lu_site *site;
358 struct vvp_pgcache_id id;
361 vvp_pgcache_id_unpack(pos, &id);
364 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
366 clob = vvp_pgcache_obj(env, dev, &id);
368 struct cl_object_header *hdr;
372 /* got an object. Find next page. */
373 hdr = cl_object_header(clob);
375 spin_lock(&hdr->coh_page_guard);
376 nr = radix_tree_gang_lookup(&hdr->coh_tree,
380 id.vpi_index = pg->cp_index;
381 /* Cant support over 16T file */
382 nr = !(pg->cp_index > 0xffffffff);
384 spin_unlock(&hdr->coh_page_guard);
386 lu_object_ref_del(&clob->co_lu, "dump", cfs_current());
387 cl_object_put(env, clob);
389 return vvp_pgcache_id_pack(&id);
391 /* to the next object. */
394 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
400 #define seq_page_flag(seq, page, flag, has_flags) do { \
401 if (test_bit(PG_##flag, &(page)->flags)) { \
402 seq_printf(seq, "%s"#flag, has_flags ? "|" : ""); \
407 static void vvp_pgcache_page_show(const struct lu_env *env,
408 struct seq_file *seq, struct cl_page *page)
410 struct ccc_page *cpg;
414 cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
415 vmpage = cpg->cpg_page;
416 seq_printf(seq," %5i | %p %p %s %s %s %s | %p %lu/%u(%p) %lu %u [",
420 cpg->cpg_write_queued ? "wq" : "- ",
421 cpg->cpg_defer_uptodate ? "du" : "- ",
422 PageWriteback(vmpage) ? "wb" : "-",
423 vmpage, vmpage->mapping->host->i_ino,
424 vmpage->mapping->host->i_generation,
425 vmpage->mapping->host, vmpage->index,
428 seq_page_flag(seq, vmpage, locked, has_flags);
429 seq_page_flag(seq, vmpage, error, has_flags);
430 seq_page_flag(seq, vmpage, referenced, has_flags);
431 seq_page_flag(seq, vmpage, uptodate, has_flags);
432 seq_page_flag(seq, vmpage, dirty, has_flags);
433 seq_page_flag(seq, vmpage, writeback, has_flags);
434 seq_printf(seq, "%s]\n", has_flags ? "" : "-");
437 static int vvp_pgcache_show(struct seq_file *f, void *v)
440 struct ll_sb_info *sbi;
441 struct cl_object *clob;
443 struct cl_page *page;
444 struct cl_object_header *hdr;
445 struct vvp_pgcache_id id;
449 env = cl_env_get(&refcheck);
452 vvp_pgcache_id_unpack(pos, &id);
454 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
456 hdr = cl_object_header(clob);
458 spin_lock(&hdr->coh_page_guard);
459 page = cl_page_lookup(hdr, id.vpi_index);
460 spin_unlock(&hdr->coh_page_guard);
462 seq_printf(f, "%8x@"DFID": ",
463 id.vpi_index, PFID(&hdr->coh_lu.loh_fid));
465 vvp_pgcache_page_show(env, f, page);
466 cl_page_put(env, page);
468 seq_puts(f, "missing\n");
469 lu_object_ref_del(&clob->co_lu, "dump", cfs_current());
470 cl_object_put(env, clob);
472 seq_printf(f, "%llx missing\n", pos);
473 cl_env_put(env, &refcheck);
476 result = PTR_ERR(env);
480 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
482 struct ll_sb_info *sbi;
488 env = cl_env_get(&refcheck);
491 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
492 pos = ERR_PTR(-EFBIG);
494 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
499 cl_env_put(env, &refcheck);
504 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
506 struct ll_sb_info *sbi;
510 env = cl_env_get(&refcheck);
513 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
516 cl_env_put(env, &refcheck);
521 static void vvp_pgcache_stop(struct seq_file *f, void *v)
526 static struct seq_operations vvp_pgcache_ops = {
527 .start = vvp_pgcache_start,
528 .next = vvp_pgcache_next,
529 .stop = vvp_pgcache_stop,
530 .show = vvp_pgcache_show
533 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
535 struct proc_dir_entry *dp = PDE(inode);
536 struct ll_sb_info *sbi = dp->data;
537 struct seq_file *seq;
540 result = seq_open(filp, &vvp_pgcache_ops);
542 seq = filp->private_data;
548 struct file_operations vvp_dump_pgcache_file_ops = {
549 .owner = THIS_MODULE,
550 .open = vvp_dump_pgcache_seq_open,
553 .release = seq_release,