lustre/llite/vvp_dev.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   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.
   9  *
  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).
  15  *
  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
  19  *
  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
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * cl_device and cl_device_type implementation for VVP layer.
  37  *
  38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39  *   Author: Jinshan Xiong <jinshan.xiong@intel.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_LLITE
  43
  44
  45 #include <obd.h>
  46 #include "llite_internal.h"
  47 #include "vvp_internal.h"
  48
  49 /*****************************************************************************
  50  *
  51  * Vvp device and device type functions.
  52  *
  53  */
  54
  55 /*
  56  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  57  * "llite_" (var. "ll_") prefix.
  58  */
  59
  60 static struct kmem_cache *vvp_thread_kmem;
  61 static struct kmem_cache *vvp_session_kmem;
  62 static struct lu_kmem_descr vvp_caches[] = {
  63         {
  64                 .ckd_cache = &vvp_thread_kmem,
  65                 .ckd_name  = "vvp_thread_kmem",
  66                 .ckd_size  = sizeof (struct vvp_thread_info),
  67         },
  68         {
  69                 .ckd_cache = &vvp_session_kmem,
  70                 .ckd_name  = "vvp_session_kmem",
  71                 .ckd_size  = sizeof (struct vvp_session)
  72         },
  73         {
  74                 .ckd_cache = NULL
  75         }
  76 };
  77
  78 static void *vvp_key_init(const struct lu_context *ctx,
  79                           struct lu_context_key *key)
  80 {
  81         struct vvp_thread_info *info;
  82
  83         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, GFP_NOFS);
  84         if (info == NULL)
  85                 info = ERR_PTR(-ENOMEM);
  86         return info;
  87 }
  88
  89 static void vvp_key_fini(const struct lu_context *ctx,
  90                          struct lu_context_key *key, void *data)
  91 {
  92         struct vvp_thread_info *info = data;
  93         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
  94 }
  95
  96 static void *vvp_session_key_init(const struct lu_context *ctx,
  97                                   struct lu_context_key *key)
  98 {
  99         struct vvp_session *session;
 100
 101         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
 102         if (session == NULL)
 103                 session = ERR_PTR(-ENOMEM);
 104         return session;
 105 }
 106
 107 static void vvp_session_key_fini(const struct lu_context *ctx,
 108                                  struct lu_context_key *key, void *data)
 109 {
 110         struct vvp_session *session = data;
 111         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 112 }
 113
 114
 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
 119 };
 120
 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
 125 };
 126
 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);
 129
 130 static const struct lu_device_operations vvp_lu_ops = {
 131         .ldo_object_alloc      = vvp_object_alloc
 132 };
 133
 134 static const struct cl_device_operations vvp_cl_ops = {
 135         .cdo_req_init = ccc_req_init
 136 };
 137
 138 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 139                                           struct lu_device_type *t,
 140                                           struct lustre_cfg *cfg)
 141 {
 142         return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
 143 }
 144
 145 static const struct lu_device_type_operations vvp_device_type_ops = {
 146         .ldto_init = vvp_type_init,
 147         .ldto_fini = vvp_type_fini,
 148
 149         .ldto_start = vvp_type_start,
 150         .ldto_stop  = vvp_type_stop,
 151
 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
 156 };
 157
 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
 163 };
 164
 165 /**
 166  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 167  * pressure, when environments cannot be allocated.
 168  */
 169 int vvp_global_init(void)
 170 {
 171         int result;
 172
 173         result = lu_kmem_init(vvp_caches);
 174         if (result == 0) {
 175                 result = ccc_global_init(&vvp_device_type);
 176                 if (result != 0)
 177                         lu_kmem_fini(vvp_caches);
 178         }
 179         return result;
 180 }
 181
 182 void vvp_global_fini(void)
 183 {
 184         ccc_global_fini(&vvp_device_type);
 185         lu_kmem_fini(vvp_caches);
 186 }
 187
 188
 189 /*****************************************************************************
 190  *
 191  * mirror obd-devices into cl devices.
 192  *
 193  */
 194
 195 int cl_sb_init(struct super_block *sb)
 196 {
 197         struct ll_sb_info *sbi;
 198         struct cl_device  *cl;
 199         struct lu_env     *env;
 200         int rc = 0;
 201         int refcheck;
 202
 203         sbi  = ll_s2sbi(sb);
 204         env = cl_env_get(&refcheck);
 205         if (!IS_ERR(env)) {
 206                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 207                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 208                 if (!IS_ERR(cl)) {
 209                         cl2ccc_dev(cl)->cdv_sb = sb;
 210                         sbi->ll_cl = cl;
 211                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 212                 }
 213                 cl_env_put(env, &refcheck);
 214         } else
 215                 rc = PTR_ERR(env);
 216         RETURN(rc);
 217 }
 218
 219 int cl_sb_fini(struct super_block *sb)
 220 {
 221         struct ll_sb_info *sbi;
 222         struct lu_env     *env;
 223         struct cl_device  *cld;
 224         int                refcheck;
 225         int                result;
 226
 227         ENTRY;
 228         sbi = ll_s2sbi(sb);
 229         env = cl_env_get(&refcheck);
 230         if (!IS_ERR(env)) {
 231                 cld = sbi->ll_cl;
 232
 233                 if (cld != NULL) {
 234                         cl_stack_fini(env, cld);
 235                         sbi->ll_cl = NULL;
 236                         sbi->ll_site = NULL;
 237                 }
 238                 cl_env_put(env, &refcheck);
 239                 result = 0;
 240         } else {
 241                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 242                 result = PTR_ERR(env);
 243         }
 244
 245         RETURN(result);
 246 }
 247
 248 /****************************************************************************
 249  *
 250  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 251  *
 252  ****************************************************************************/
 253
 254 /*
 255  * To represent contents of a page cache as a byte stream, following
 256  * information if encoded in 64bit offset:
 257  *
 258  *       - file hash bucket in lu_site::ls_hash[]       28bits
 259  *
 260  *       - how far file is from bucket head              4bits
 261  *
 262  *       - page index                                   32bits
 263  *
 264  * First two data identify a file in the cache uniquely.
 265  */
 266
 267 #define PGC_OBJ_SHIFT (32 + 4)
 268 #define PGC_DEPTH_SHIFT (32)
 269
 270 struct vvp_pgcache_id {
 271         unsigned                 vpi_bucket;
 272         unsigned                 vpi_depth;
 273         uint32_t                 vpi_index;
 274
 275         unsigned                 vpi_curdep;
 276         struct lu_object_header *vpi_obj;
 277 };
 278
 279 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 280 {
 281         CLASSERT(sizeof(pos) == sizeof(__u64));
 282
 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);
 286 }
 287
 288 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 289 {
 290         return
 291                 ((__u64)id->vpi_index) |
 292                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 293                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 294 }
 295
 296 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 297                                struct hlist_node *hnode, void *data)
 298 {
 299         struct vvp_pgcache_id   *id  = data;
 300         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 301
 302         if (id->vpi_curdep-- > 0)
 303                 return 0; /* continue */
 304
 305         if (lu_object_is_dying(hdr))
 306                 return 1;
 307
 308         cfs_hash_get(hs, hnode);
 309         id->vpi_obj = hdr;
 310         return 1;
 311 }
 312
 313 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 314                                          struct lu_device *dev,
 315                                          struct vvp_pgcache_id *id)
 316 {
 317         LASSERT(lu_device_is_cl(dev));
 318
 319         id->vpi_depth &= 0xf;
 320         id->vpi_obj    = NULL;
 321         id->vpi_curdep = id->vpi_depth;
 322
 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;
 327
 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);
 332                 }
 333                 lu_object_put(env, lu_object_top(id->vpi_obj));
 334
 335         } else if (id->vpi_curdep > 0) {
 336                 id->vpi_depth = 0xf;
 337         }
 338         return NULL;
 339 }
 340
 341 static loff_t vvp_pgcache_find(const struct lu_env *env,
 342                                struct lu_device *dev, loff_t pos)
 343 {
 344         struct cl_object     *clob;
 345         struct lu_site       *site;
 346         struct vvp_pgcache_id id;
 347
 348         site = dev->ld_site;
 349         vvp_pgcache_id_unpack(pos, &id);
 350
 351         while (1) {
 352                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 353                         return ~0ULL;
 354                 clob = vvp_pgcache_obj(env, dev, &id);
 355                 if (clob != NULL) {
 356                         struct inode *inode = ccc_object_inode(clob);
 357                         struct page *vmpage;
 358                         int nr;
 359
 360                         nr = find_get_pages_contig(inode->i_mapping,
 361                                                    id.vpi_index, 1, &vmpage);
 362                         if (nr > 0) {
 363                                 id.vpi_index = vmpage->index;
 364                                 /* Cant support over 16T file */
 365                                 nr = !(vmpage->index > 0xffffffff);
 366                                 page_cache_release(vmpage);
 367                         }
 368
 369                         lu_object_ref_del(&clob->co_lu, "dump", current);
 370                         cl_object_put(env, clob);
 371                         if (nr > 0)
 372                                 return vvp_pgcache_id_pack(&id);
 373                 }
 374                 /* to the next object. */
 375                 ++id.vpi_depth;
 376                 id.vpi_depth &= 0xf;
 377                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 378                         return ~0ULL;
 379                 id.vpi_index = 0;
 380         }
 381 }
 382
 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 ? "|" : "");       \
 386                 has_flags = 1;                                          \
 387         }                                                               \
 388 } while(0)
 389
 390 static void vvp_pgcache_page_show(const struct lu_env *env,
 391                                   struct seq_file *seq, struct cl_page *page)
 392 {
 393         struct ccc_page *cpg;
 394         struct page      *vmpage;
 395         int              has_flags;
 396
 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 [",
 400                    0 /* gen */,
 401                    cpg, page,
 402                    "none",
 403                    cpg->cpg_write_queued ? "wq" : "- ",
 404                    cpg->cpg_defer_uptodate ? "du" : "- ",
 405                    PageWriteback(vmpage) ? "wb" : "-",
 406                    vmpage,
 407                    PFID(ll_inode2fid(vmpage->mapping->host)),
 408                    vmpage->mapping->host, vmpage->index,
 409                    page_count(vmpage));
 410         has_flags = 0;
 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 ? "" : "-");
 418 }
 419
 420 static int vvp_pgcache_show(struct seq_file *f, void *v)
 421 {
 422         loff_t                   pos;
 423         struct ll_sb_info       *sbi;
 424         struct cl_object        *clob;
 425         struct lu_env           *env;
 426         struct vvp_pgcache_id    id;
 427         int                      refcheck;
 428         int                      result;
 429
 430         env = cl_env_get(&refcheck);
 431         if (!IS_ERR(env)) {
 432                 pos = *(loff_t *) v;
 433                 vvp_pgcache_id_unpack(pos, &id);
 434                 sbi = f->private;
 435                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 436                 if (clob != NULL) {
 437                         struct inode *inode = ccc_object_inode(clob);
 438                         struct cl_page *page = NULL;
 439                         struct page *vmpage;
 440
 441                         result = find_get_pages_contig(inode->i_mapping,
 442                                                       id.vpi_index, 1, &vmpage);
 443                         if (result > 0) {
 444                                 lock_page(vmpage);
 445                                 page = cl_vmpage_page(vmpage, clob);
 446                                 unlock_page(vmpage);
 447
 448                                 page_cache_release(vmpage);
 449                         }
 450
 451                         seq_printf(f, "%8x@"DFID": ", id.vpi_index,
 452                                    PFID(lu_object_fid(&clob->co_lu)));
 453                         if (page != NULL) {
 454                                 vvp_pgcache_page_show(env, f, page);
 455                                 cl_page_put(env, page);
 456                         } else
 457                                 seq_puts(f, "missing\n");
 458                         lu_object_ref_del(&clob->co_lu, "dump", current);
 459                         cl_object_put(env, clob);
 460                 } else
 461                         seq_printf(f, "%llx missing\n", pos);
 462                 cl_env_put(env, &refcheck);
 463                 result = 0;
 464         } else
 465                 result = PTR_ERR(env);
 466         return result;
 467 }
 468
 469 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 470 {
 471         struct ll_sb_info *sbi;
 472         struct lu_env     *env;
 473         int                refcheck;
 474
 475         sbi = f->private;
 476
 477         env = cl_env_get(&refcheck);
 478         if (!IS_ERR(env)) {
 479                 sbi = f->private;
 480                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 481                         pos = ERR_PTR(-EFBIG);
 482                 else {
 483                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 484                                                 *pos);
 485                         if (*pos == ~0ULL)
 486                                 pos = NULL;
 487                 }
 488                 cl_env_put(env, &refcheck);
 489         }
 490         return pos;
 491 }
 492
 493 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 494 {
 495         struct ll_sb_info *sbi;
 496         struct lu_env     *env;
 497         int                refcheck;
 498
 499         env = cl_env_get(&refcheck);
 500         if (!IS_ERR(env)) {
 501                 sbi = f->private;
 502                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 503                 if (*pos == ~0ULL)
 504                         pos = NULL;
 505                 cl_env_put(env, &refcheck);
 506         }
 507         return pos;
 508 }
 509
 510 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 511 {
 512         /* Nothing to do */
 513 }
 514
 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
 520 };
 521
 522 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 523 {
 524         struct ll_sb_info       *sbi = PDE_DATA(inode);
 525         struct seq_file         *seq;
 526         int                     result;
 527
 528         result = seq_open(filp, &vvp_pgcache_ops);
 529         if (result == 0) {
 530                 seq = filp->private_data;
 531                 seq->private = sbi;
 532         }
 533         return result;
 534 }
 535
 536 const struct file_operations vvp_dump_pgcache_file_ops = {
 537         .owner   = THIS_MODULE,
 538         .open    = vvp_dump_pgcache_seq_open,
 539         .read    = seq_read,
 540         .llseek  = seq_lseek,
 541         .release = seq_release,
 542 };