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 /*
  31  * This file is part of Lustre, http://www.lustre.org/
  32  * Lustre is a trademark of Sun Microsystems, Inc.
  33  *
  34  * cl_device and cl_device_type implementation for VVP layer.
  35  *
  36  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  37  */
  38
  39 #define DEBUG_SUBSYSTEM S_LLITE
  40
  41 #ifndef __KERNEL__
  42 # error This file is kernel only.
  43 #endif
  44
  45 #include <obd.h>
  46 #include <lustre_lite.h>
  47
  48 #include "vvp_internal.h"
  49
  50 /*****************************************************************************
  51  *
  52  * Vvp device and device type functions.
  53  *
  54  */
  55
  56 /*
  57  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  58  * "llite_" (var. "ll_") prefix.
  59  */
  60
  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[] = {
  65         {
  66                 .ckd_cache = &vvp_page_kmem,
  67                 .ckd_name  = "vvp_page_kmem",
  68                 .ckd_size  = sizeof (struct ccc_page)
  69         },
  70         {
  71                 .ckd_cache = &vvp_thread_kmem,
  72                 .ckd_name  = "vvp_thread_kmem",
  73                 .ckd_size  = sizeof (struct vvp_thread_info),
  74         },
  75         {
  76                 .ckd_cache = &vvp_session_kmem,
  77                 .ckd_name  = "vvp_session_kmem",
  78                 .ckd_size  = sizeof (struct vvp_session)
  79         },
  80         {
  81                 .ckd_cache = NULL
  82         }
  83 };
  84
  85 static void *vvp_key_init(const struct lu_context *ctx,
  86                           struct lu_context_key *key)
  87 {
  88         struct vvp_thread_info *info;
  89
  90         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, CFS_ALLOC_IO);
  91         if (info == NULL)
  92                 info = ERR_PTR(-ENOMEM);
  93         return info;
  94 }
  95
  96 static void vvp_key_fini(const struct lu_context *ctx,
  97                          struct lu_context_key *key, void *data)
  98 {
  99         struct vvp_thread_info *info = data;
 100         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
 101 }
 102
 103 static void *vvp_session_key_init(const struct lu_context *ctx,
 104                                   struct lu_context_key *key)
 105 {
 106         struct vvp_session *session;
 107
 108         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, CFS_ALLOC_IO);
 109         if (session == NULL)
 110                 session = ERR_PTR(-ENOMEM);
 111         return session;
 112 }
 113
 114 static void vvp_session_key_fini(const struct lu_context *ctx,
 115                                  struct lu_context_key *key, void *data)
 116 {
 117         struct vvp_session *session = data;
 118         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 119 }
 120
 121
 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
 126 };
 127
 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
 132 };
 133
 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);
 136
 137 static const struct lu_device_operations vvp_lu_ops = {
 138         .ldo_object_alloc      = vvp_object_alloc
 139 };
 140
 141 static const struct cl_device_operations vvp_cl_ops = {
 142         .cdo_req_init = ccc_req_init
 143 };
 144
 145 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 146                                           struct lu_device_type *t,
 147                                           struct lustre_cfg *cfg)
 148 {
 149         return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
 150 }
 151
 152 static const struct lu_device_type_operations vvp_device_type_ops = {
 153         .ldto_init = vvp_type_init,
 154         .ldto_fini = vvp_type_fini,
 155
 156         .ldto_start = vvp_type_start,
 157         .ldto_stop  = vvp_type_stop,
 158
 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
 163 };
 164
 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
 170 };
 171
 172 /**
 173  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 174  * pressure, when environments cannot be allocated.
 175  */
 176 int vvp_global_init(void)
 177 {
 178         int result;
 179
 180         result = lu_kmem_init(vvp_caches);
 181         if (result == 0) {
 182                 result = ccc_global_init(&vvp_device_type);
 183                 if (result != 0)
 184                         lu_kmem_fini(vvp_caches);
 185         }
 186         return result;
 187 }
 188
 189 void vvp_global_fini(void)
 190 {
 191         ccc_global_fini(&vvp_device_type);
 192         lu_kmem_fini(vvp_caches);
 193 }
 194
 195
 196 /*****************************************************************************
 197  *
 198  * mirror obd-devices into cl devices.
 199  *
 200  */
 201
 202 int cl_sb_init(struct super_block *sb)
 203 {
 204         struct ll_sb_info *sbi;
 205         struct cl_device  *cl;
 206         struct lu_env     *env;
 207         int rc = 0;
 208         int refcheck;
 209
 210         sbi  = ll_s2sbi(sb);
 211         env = cl_env_get(&refcheck);
 212         if (!IS_ERR(env)) {
 213                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 214                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 215                 if (!IS_ERR(cl)) {
 216                         cl2ccc_dev(cl)->cdv_sb = sb;
 217                         sbi->ll_cl = cl;
 218                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 219                 }
 220                 cl_env_put(env, &refcheck);
 221         } else
 222                 rc = PTR_ERR(env);
 223         RETURN(rc);
 224 }
 225
 226 int cl_sb_fini(struct super_block *sb)
 227 {
 228         struct ll_sb_info *sbi;
 229         struct lu_env     *env;
 230         struct cl_device  *cld;
 231         int                refcheck;
 232         int                result;
 233
 234         ENTRY;
 235         sbi = ll_s2sbi(sb);
 236         env = cl_env_get(&refcheck);
 237         if (!IS_ERR(env)) {
 238                 cld = sbi->ll_cl;
 239
 240                 if (cld != NULL) {
 241                         cl_stack_fini(env, cld);
 242                         sbi->ll_cl = NULL;
 243                         sbi->ll_site = NULL;
 244                 }
 245                 cl_env_put(env, &refcheck);
 246                 result = 0;
 247         } else {
 248                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 249                 result = PTR_ERR(env);
 250         }
 251         /*
 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).
 255          */
 256         lu_types_stop();
 257         RETURN(result);
 258 }
 259
 260 /****************************************************************************
 261  *
 262  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 263  *
 264  ****************************************************************************/
 265
 266 /*
 267  * To represent contents of a page cache as a byte stream, following
 268  * information if encoded in 64bit offset:
 269  *
 270  *       - file hash bucket in lu_site::ls_hash[]       28bits
 271  *
 272  *       - how far file is from bucket head              4bits
 273  *
 274  *       - page index                                   32bits
 275  *
 276  * First two data identify a file in the cache uniquely.
 277  */
 278
 279 #define PGC_OBJ_SHIFT (32 + 4)
 280 #define PGC_DEPTH_SHIFT (32)
 281
 282 struct vvp_pgcache_id {
 283         unsigned                 vpi_bucket;
 284         unsigned                 vpi_depth;
 285         uint32_t                 vpi_index;
 286
 287         unsigned                 vpi_curdep;
 288         struct lu_object_header *vpi_obj;
 289 };
 290
 291 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 292 {
 293         CLASSERT(sizeof(pos) == sizeof(__u64));
 294
 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);
 298 }
 299
 300 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 301 {
 302         return
 303                 ((__u64)id->vpi_index) |
 304                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 305                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 306 }
 307
 308 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 309                                cfs_hlist_node_t *hnode, void *data)
 310 {
 311         struct vvp_pgcache_id   *id  = data;
 312         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 313
 314         if (id->vpi_curdep-- > 0)
 315                 return 0; /* continue */
 316
 317         if (lu_object_is_dying(hdr))
 318                 return 1;
 319
 320         cfs_hash_get(hs, hnode);
 321         id->vpi_obj = hdr;
 322         return 1;
 323 }
 324
 325 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 326                                          struct lu_device *dev,
 327                                          struct vvp_pgcache_id *id)
 328 {
 329         LASSERT(lu_device_is_cl(dev));
 330
 331         id->vpi_depth &= 0xf;
 332         id->vpi_obj    = NULL;
 333         id->vpi_curdep = id->vpi_depth;
 334
 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;
 339
 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);
 344                 }
 345                 lu_object_put(env, lu_object_top(id->vpi_obj));
 346
 347         } else if (id->vpi_curdep > 0) {
 348                 id->vpi_depth = 0xf;
 349         }
 350         return NULL;
 351 }
 352
 353 static loff_t vvp_pgcache_find(const struct lu_env *env,
 354                                struct lu_device *dev, loff_t pos)
 355 {
 356         struct cl_object     *clob;
 357         struct lu_site       *site;
 358         struct vvp_pgcache_id id;
 359
 360         site = dev->ld_site;
 361         vvp_pgcache_id_unpack(pos, &id);
 362
 363         while (1) {
 364                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 365                         return ~0ULL;
 366                 clob = vvp_pgcache_obj(env, dev, &id);
 367                 if (clob != NULL) {
 368                         struct cl_object_header *hdr;
 369                         int                      nr;
 370                         struct cl_page          *pg;
 371
 372                         /* got an object. Find next page. */
 373                         hdr = cl_object_header(clob);
 374
 375                         spin_lock(&hdr->coh_page_guard);
 376                         nr = radix_tree_gang_lookup(&hdr->coh_tree,
 377                                                     (void **)&pg,
 378                                                     id.vpi_index, 1);
 379                         if (nr > 0) {
 380                                 id.vpi_index = pg->cp_index;
 381                                 /* Cant support over 16T file */
 382                                 nr = !(pg->cp_index > 0xffffffff);
 383                         }
 384                         spin_unlock(&hdr->coh_page_guard);
 385
 386                         lu_object_ref_del(&clob->co_lu, "dump", cfs_current());
 387                         cl_object_put(env, clob);
 388                         if (nr > 0)
 389                                 return vvp_pgcache_id_pack(&id);
 390                 }
 391                 /* to the next object. */
 392                 ++id.vpi_depth;
 393                 id.vpi_depth &= 0xf;
 394                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 395                         return ~0ULL;
 396                 id.vpi_index = 0;
 397         }
 398 }
 399
 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 ? "|" : "");       \
 403                 has_flags = 1;                                          \
 404         }                                                               \
 405 } while(0)
 406
 407 static void vvp_pgcache_page_show(const struct lu_env *env,
 408                                   struct seq_file *seq, struct cl_page *page)
 409 {
 410         struct ccc_page *cpg;
 411         cfs_page_t      *vmpage;
 412         int              has_flags;
 413
 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 [",
 417                    0 /* gen */,
 418                    cpg, page,
 419                    "none",
 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,
 426                    page_count(vmpage));
 427         has_flags = 0;
 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 ? "" : "-");
 435 }
 436
 437 static int vvp_pgcache_show(struct seq_file *f, void *v)
 438 {
 439         loff_t                   pos;
 440         struct ll_sb_info       *sbi;
 441         struct cl_object        *clob;
 442         struct lu_env           *env;
 443         struct cl_page          *page;
 444         struct cl_object_header *hdr;
 445         struct vvp_pgcache_id    id;
 446         int                      refcheck;
 447         int                      result;
 448
 449         env = cl_env_get(&refcheck);
 450         if (!IS_ERR(env)) {
 451                 pos = *(loff_t *) v;
 452                 vvp_pgcache_id_unpack(pos, &id);
 453                 sbi = f->private;
 454                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 455                 if (clob != NULL) {
 456                         hdr = cl_object_header(clob);
 457
 458                         spin_lock(&hdr->coh_page_guard);
 459                         page = cl_page_lookup(hdr, id.vpi_index);
 460                         spin_unlock(&hdr->coh_page_guard);
 461
 462                         seq_printf(f, "%8x@"DFID": ",
 463                                    id.vpi_index, PFID(&hdr->coh_lu.loh_fid));
 464                         if (page != NULL) {
 465                                 vvp_pgcache_page_show(env, f, page);
 466                                 cl_page_put(env, page);
 467                         } else
 468                                 seq_puts(f, "missing\n");
 469                         lu_object_ref_del(&clob->co_lu, "dump", cfs_current());
 470                         cl_object_put(env, clob);
 471                 } else
 472                         seq_printf(f, "%llx missing\n", pos);
 473                 cl_env_put(env, &refcheck);
 474                 result = 0;
 475         } else
 476                 result = PTR_ERR(env);
 477         return result;
 478 }
 479
 480 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 481 {
 482         struct ll_sb_info *sbi;
 483         struct lu_env     *env;
 484         int                refcheck;
 485
 486         sbi = f->private;
 487
 488         env = cl_env_get(&refcheck);
 489         if (!IS_ERR(env)) {
 490                 sbi = f->private;
 491                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 492                         pos = ERR_PTR(-EFBIG);
 493                 else {
 494                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 495                                                 *pos);
 496                         if (*pos == ~0ULL)
 497                                 pos = NULL;
 498                 }
 499                 cl_env_put(env, &refcheck);
 500         }
 501         return pos;
 502 }
 503
 504 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 505 {
 506         struct ll_sb_info *sbi;
 507         struct lu_env     *env;
 508         int                refcheck;
 509
 510         env = cl_env_get(&refcheck);
 511         if (!IS_ERR(env)) {
 512                 sbi = f->private;
 513                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 514                 if (*pos == ~0ULL)
 515                         pos = NULL;
 516                 cl_env_put(env, &refcheck);
 517         }
 518         return pos;
 519 }
 520
 521 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 522 {
 523         /* Nothing to do */
 524 }
 525
 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
 531 };
 532
 533 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 534 {
 535         struct proc_dir_entry *dp  = PDE(inode);
 536         struct ll_sb_info     *sbi = dp->data;
 537         struct seq_file       *seq;
 538         int                    result;
 539
 540         result = seq_open(filp, &vvp_pgcache_ops);
 541         if (result == 0) {
 542                 seq = filp->private_data;
 543                 seq->private = sbi;
 544         }
 545         return result;
 546 }
 547
 548 struct file_operations vvp_dump_pgcache_file_ops = {
 549         .owner   = THIS_MODULE,
 550         .open    = vvp_dump_pgcache_seq_open,
 551         .read    = seq_read,
 552         .llseek  = seq_lseek,
 553         .release = seq_release,
 554 };