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, 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  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #ifndef __KERNEL__
  44 # error This file is kernel only.
  45 #endif
  46
  47 #include <obd.h>
  48 #include <lustre_lite.h>
  49
  50 #include "vvp_internal.h"
  51
  52 /*****************************************************************************
  53  *
  54  * Vvp device and device type functions.
  55  *
  56  */
  57
  58 /*
  59  * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
  60  * "llite_" (var. "ll_") prefix.
  61  */
  62
  63 cfs_mem_cache_t *vvp_thread_kmem;
  64 static cfs_mem_cache_t *vvp_session_kmem;
  65 static struct lu_kmem_descr vvp_caches[] = {
  66         {
  67                 .ckd_cache = &vvp_thread_kmem,
  68                 .ckd_name  = "vvp_thread_kmem",
  69                 .ckd_size  = sizeof (struct vvp_thread_info),
  70         },
  71         {
  72                 .ckd_cache = &vvp_session_kmem,
  73                 .ckd_name  = "vvp_session_kmem",
  74                 .ckd_size  = sizeof (struct vvp_session)
  75         },
  76         {
  77                 .ckd_cache = NULL
  78         }
  79 };
  80
  81 static void *vvp_key_init(const struct lu_context *ctx,
  82                           struct lu_context_key *key)
  83 {
  84         struct vvp_thread_info *info;
  85
  86         OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, CFS_ALLOC_IO);
  87         if (info == NULL)
  88                 info = ERR_PTR(-ENOMEM);
  89         return info;
  90 }
  91
  92 static void vvp_key_fini(const struct lu_context *ctx,
  93                          struct lu_context_key *key, void *data)
  94 {
  95         struct vvp_thread_info *info = data;
  96         OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
  97 }
  98
  99 static void *vvp_session_key_init(const struct lu_context *ctx,
 100                                   struct lu_context_key *key)
 101 {
 102         struct vvp_session *session;
 103
 104         OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, CFS_ALLOC_IO);
 105         if (session == NULL)
 106                 session = ERR_PTR(-ENOMEM);
 107         return session;
 108 }
 109
 110 static void vvp_session_key_fini(const struct lu_context *ctx,
 111                                  struct lu_context_key *key, void *data)
 112 {
 113         struct vvp_session *session = data;
 114         OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
 115 }
 116
 117
 118 struct lu_context_key vvp_key = {
 119         .lct_tags = LCT_CL_THREAD,
 120         .lct_init = vvp_key_init,
 121         .lct_fini = vvp_key_fini
 122 };
 123
 124 struct lu_context_key vvp_session_key = {
 125         .lct_tags = LCT_SESSION,
 126         .lct_init = vvp_session_key_init,
 127         .lct_fini = vvp_session_key_fini
 128 };
 129
 130 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
 131 LU_TYPE_INIT_FINI(vvp, &ccc_key, &ccc_session_key, &vvp_key, &vvp_session_key);
 132
 133 static const struct lu_device_operations vvp_lu_ops = {
 134         .ldo_object_alloc      = vvp_object_alloc
 135 };
 136
 137 static const struct cl_device_operations vvp_cl_ops = {
 138         .cdo_req_init = ccc_req_init
 139 };
 140
 141 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
 142                                           struct lu_device_type *t,
 143                                           struct lustre_cfg *cfg)
 144 {
 145         return ccc_device_alloc(env, t, cfg, &vvp_lu_ops, &vvp_cl_ops);
 146 }
 147
 148 static const struct lu_device_type_operations vvp_device_type_ops = {
 149         .ldto_init = vvp_type_init,
 150         .ldto_fini = vvp_type_fini,
 151
 152         .ldto_start = vvp_type_start,
 153         .ldto_stop  = vvp_type_stop,
 154
 155         .ldto_device_alloc = vvp_device_alloc,
 156         .ldto_device_free  = ccc_device_free,
 157         .ldto_device_init  = ccc_device_init,
 158         .ldto_device_fini  = ccc_device_fini
 159 };
 160
 161 struct lu_device_type vvp_device_type = {
 162         .ldt_tags     = LU_DEVICE_CL,
 163         .ldt_name     = LUSTRE_VVP_NAME,
 164         .ldt_ops      = &vvp_device_type_ops,
 165         .ldt_ctx_tags = LCT_CL_THREAD
 166 };
 167
 168 /**
 169  * A mutex serializing calls to vvp_inode_fini() under extreme memory
 170  * pressure, when environments cannot be allocated.
 171  */
 172 int vvp_global_init(void)
 173 {
 174         int result;
 175
 176         result = lu_kmem_init(vvp_caches);
 177         if (result == 0) {
 178                 result = ccc_global_init(&vvp_device_type);
 179                 if (result != 0)
 180                         lu_kmem_fini(vvp_caches);
 181         }
 182         return result;
 183 }
 184
 185 void vvp_global_fini(void)
 186 {
 187         ccc_global_fini(&vvp_device_type);
 188         lu_kmem_fini(vvp_caches);
 189 }
 190
 191
 192 /*****************************************************************************
 193  *
 194  * mirror obd-devices into cl devices.
 195  *
 196  */
 197
 198 int cl_sb_init(struct super_block *sb)
 199 {
 200         struct ll_sb_info *sbi;
 201         struct cl_device  *cl;
 202         struct lu_env     *env;
 203         int rc = 0;
 204         int refcheck;
 205
 206         sbi  = ll_s2sbi(sb);
 207         env = cl_env_get(&refcheck);
 208         if (!IS_ERR(env)) {
 209                 cl = cl_type_setup(env, NULL, &vvp_device_type,
 210                                    sbi->ll_dt_exp->exp_obd->obd_lu_dev);
 211                 if (!IS_ERR(cl)) {
 212                         cl2ccc_dev(cl)->cdv_sb = sb;
 213                         sbi->ll_cl = cl;
 214                         sbi->ll_site = cl2lu_dev(cl)->ld_site;
 215                 }
 216                 cl_env_put(env, &refcheck);
 217         } else
 218                 rc = PTR_ERR(env);
 219         RETURN(rc);
 220 }
 221
 222 int cl_sb_fini(struct super_block *sb)
 223 {
 224         struct ll_sb_info *sbi;
 225         struct lu_env     *env;
 226         struct cl_device  *cld;
 227         int                refcheck;
 228         int                result;
 229
 230         ENTRY;
 231         sbi = ll_s2sbi(sb);
 232         env = cl_env_get(&refcheck);
 233         if (!IS_ERR(env)) {
 234                 cld = sbi->ll_cl;
 235
 236                 if (cld != NULL) {
 237                         cl_stack_fini(env, cld);
 238                         sbi->ll_cl = NULL;
 239                         sbi->ll_site = NULL;
 240                 }
 241                 cl_env_put(env, &refcheck);
 242                 result = 0;
 243         } else {
 244                 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
 245                 result = PTR_ERR(env);
 246         }
 247         /*
 248          * If mount failed (sbi->ll_cl == NULL), and this there are no other
 249          * mounts, stop device types manually (this usually happens
 250          * automatically when last device is destroyed).
 251          */
 252         lu_types_stop();
 253         RETURN(result);
 254 }
 255
 256 /****************************************************************************
 257  *
 258  * /proc/fs/lustre/llite/$MNT/dump_page_cache
 259  *
 260  ****************************************************************************/
 261
 262 /*
 263  * To represent contents of a page cache as a byte stream, following
 264  * information if encoded in 64bit offset:
 265  *
 266  *       - file hash bucket in lu_site::ls_hash[]       28bits
 267  *
 268  *       - how far file is from bucket head              4bits
 269  *
 270  *       - page index                                   32bits
 271  *
 272  * First two data identify a file in the cache uniquely.
 273  */
 274
 275 #define PGC_OBJ_SHIFT (32 + 4)
 276 #define PGC_DEPTH_SHIFT (32)
 277
 278 struct vvp_pgcache_id {
 279         unsigned                 vpi_bucket;
 280         unsigned                 vpi_depth;
 281         uint32_t                 vpi_index;
 282
 283         unsigned                 vpi_curdep;
 284         struct lu_object_header *vpi_obj;
 285 };
 286
 287 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
 288 {
 289         CLASSERT(sizeof(pos) == sizeof(__u64));
 290
 291         id->vpi_index  = pos & 0xffffffff;
 292         id->vpi_depth  = (pos >> PGC_DEPTH_SHIFT) & 0xf;
 293         id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
 294 }
 295
 296 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
 297 {
 298         return
 299                 ((__u64)id->vpi_index) |
 300                 ((__u64)id->vpi_depth  << PGC_DEPTH_SHIFT) |
 301                 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
 302 }
 303
 304 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 305                                cfs_hlist_node_t *hnode, void *data)
 306 {
 307         struct vvp_pgcache_id   *id  = data;
 308         struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
 309
 310         if (id->vpi_curdep-- > 0)
 311                 return 0; /* continue */
 312
 313         if (lu_object_is_dying(hdr))
 314                 return 1;
 315
 316         cfs_hash_get(hs, hnode);
 317         id->vpi_obj = hdr;
 318         return 1;
 319 }
 320
 321 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
 322                                          struct lu_device *dev,
 323                                          struct vvp_pgcache_id *id)
 324 {
 325         LASSERT(lu_device_is_cl(dev));
 326
 327         id->vpi_depth &= 0xf;
 328         id->vpi_obj    = NULL;
 329         id->vpi_curdep = id->vpi_depth;
 330
 331         cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
 332                                 vvp_pgcache_obj_get, id);
 333         if (id->vpi_obj != NULL) {
 334                 struct lu_object *lu_obj;
 335
 336                 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
 337                 if (lu_obj != NULL) {
 338                         lu_object_ref_add(lu_obj, "dump", cfs_current());
 339                         return lu2cl(lu_obj);
 340                 }
 341                 lu_object_put(env, lu_object_top(id->vpi_obj));
 342
 343         } else if (id->vpi_curdep > 0) {
 344                 id->vpi_depth = 0xf;
 345         }
 346         return NULL;
 347 }
 348
 349 static loff_t vvp_pgcache_find(const struct lu_env *env,
 350                                struct lu_device *dev, loff_t pos)
 351 {
 352         struct cl_object     *clob;
 353         struct lu_site       *site;
 354         struct vvp_pgcache_id id;
 355
 356         site = dev->ld_site;
 357         vvp_pgcache_id_unpack(pos, &id);
 358
 359         while (1) {
 360                 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
 361                         return ~0ULL;
 362                 clob = vvp_pgcache_obj(env, dev, &id);
 363                 if (clob != NULL) {
 364                         struct cl_object_header *hdr;
 365                         int                      nr;
 366                         struct cl_page          *pg;
 367
 368                         /* got an object. Find next page. */
 369                         hdr = cl_object_header(clob);
 370
 371                         spin_lock(&hdr->coh_page_guard);
 372                         nr = radix_tree_gang_lookup(&hdr->coh_tree,
 373                                                     (void **)&pg,
 374                                                     id.vpi_index, 1);
 375                         if (nr > 0) {
 376                                 id.vpi_index = pg->cp_index;
 377                                 /* Cant support over 16T file */
 378                                 nr = !(pg->cp_index > 0xffffffff);
 379                         }
 380                         spin_unlock(&hdr->coh_page_guard);
 381
 382                         lu_object_ref_del(&clob->co_lu, "dump", cfs_current());
 383                         cl_object_put(env, clob);
 384                         if (nr > 0)
 385                                 return vvp_pgcache_id_pack(&id);
 386                 }
 387                 /* to the next object. */
 388                 ++id.vpi_depth;
 389                 id.vpi_depth &= 0xf;
 390                 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
 391                         return ~0ULL;
 392                 id.vpi_index = 0;
 393         }
 394 }
 395
 396 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 397         if (test_bit(PG_##flag, &(page)->flags)) {                  \
 398                 seq_printf(seq, "%s"#flag, has_flags ? "|" : "");       \
 399                 has_flags = 1;                                          \
 400         }                                                               \
 401 } while(0)
 402
 403 static void vvp_pgcache_page_show(const struct lu_env *env,
 404                                   struct seq_file *seq, struct cl_page *page)
 405 {
 406         struct ccc_page *cpg;
 407         cfs_page_t      *vmpage;
 408         int              has_flags;
 409
 410         cpg = cl2ccc_page(cl_page_at(page, &vvp_device_type));
 411         vmpage = cpg->cpg_page;
 412         seq_printf(seq," %5i | %p %p %s %s %s %s | %p %lu/%u(%p) %lu %u [",
 413                    0 /* gen */,
 414                    cpg, page,
 415                    "none",
 416                    cpg->cpg_write_queued ? "wq" : "- ",
 417                    cpg->cpg_defer_uptodate ? "du" : "- ",
 418                    PageWriteback(vmpage) ? "wb" : "-",
 419                    vmpage, vmpage->mapping->host->i_ino,
 420                    vmpage->mapping->host->i_generation,
 421                    vmpage->mapping->host, vmpage->index,
 422                    page_count(vmpage));
 423         has_flags = 0;
 424         seq_page_flag(seq, vmpage, locked, has_flags);
 425         seq_page_flag(seq, vmpage, error, has_flags);
 426         seq_page_flag(seq, vmpage, referenced, has_flags);
 427         seq_page_flag(seq, vmpage, uptodate, has_flags);
 428         seq_page_flag(seq, vmpage, dirty, has_flags);
 429         seq_page_flag(seq, vmpage, writeback, has_flags);
 430         seq_printf(seq, "%s]\n", has_flags ? "" : "-");
 431 }
 432
 433 static int vvp_pgcache_show(struct seq_file *f, void *v)
 434 {
 435         loff_t                   pos;
 436         struct ll_sb_info       *sbi;
 437         struct cl_object        *clob;
 438         struct lu_env           *env;
 439         struct cl_page          *page;
 440         struct cl_object_header *hdr;
 441         struct vvp_pgcache_id    id;
 442         int                      refcheck;
 443         int                      result;
 444
 445         env = cl_env_get(&refcheck);
 446         if (!IS_ERR(env)) {
 447                 pos = *(loff_t *) v;
 448                 vvp_pgcache_id_unpack(pos, &id);
 449                 sbi = f->private;
 450                 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
 451                 if (clob != NULL) {
 452                         hdr = cl_object_header(clob);
 453
 454                         spin_lock(&hdr->coh_page_guard);
 455                         page = cl_page_lookup(hdr, id.vpi_index);
 456                         spin_unlock(&hdr->coh_page_guard);
 457
 458                         seq_printf(f, "%8x@"DFID": ",
 459                                    id.vpi_index, PFID(&hdr->coh_lu.loh_fid));
 460                         if (page != NULL) {
 461                                 vvp_pgcache_page_show(env, f, page);
 462                                 cl_page_put(env, page);
 463                         } else
 464                                 seq_puts(f, "missing\n");
 465                         lu_object_ref_del(&clob->co_lu, "dump", cfs_current());
 466                         cl_object_put(env, clob);
 467                 } else
 468                         seq_printf(f, "%llx missing\n", pos);
 469                 cl_env_put(env, &refcheck);
 470                 result = 0;
 471         } else
 472                 result = PTR_ERR(env);
 473         return result;
 474 }
 475
 476 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
 477 {
 478         struct ll_sb_info *sbi;
 479         struct lu_env     *env;
 480         int                refcheck;
 481
 482         sbi = f->private;
 483
 484         env = cl_env_get(&refcheck);
 485         if (!IS_ERR(env)) {
 486                 sbi = f->private;
 487                 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
 488                         pos = ERR_PTR(-EFBIG);
 489                 else {
 490                         *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
 491                                                 *pos);
 492                         if (*pos == ~0ULL)
 493                                 pos = NULL;
 494                 }
 495                 cl_env_put(env, &refcheck);
 496         }
 497         return pos;
 498 }
 499
 500 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
 501 {
 502         struct ll_sb_info *sbi;
 503         struct lu_env     *env;
 504         int                refcheck;
 505
 506         env = cl_env_get(&refcheck);
 507         if (!IS_ERR(env)) {
 508                 sbi = f->private;
 509                 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
 510                 if (*pos == ~0ULL)
 511                         pos = NULL;
 512                 cl_env_put(env, &refcheck);
 513         }
 514         return pos;
 515 }
 516
 517 static void vvp_pgcache_stop(struct seq_file *f, void *v)
 518 {
 519         /* Nothing to do */
 520 }
 521
 522 static struct seq_operations vvp_pgcache_ops = {
 523         .start = vvp_pgcache_start,
 524         .next  = vvp_pgcache_next,
 525         .stop  = vvp_pgcache_stop,
 526         .show  = vvp_pgcache_show
 527 };
 528
 529 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
 530 {
 531         struct proc_dir_entry *dp  = PDE(inode);
 532         struct ll_sb_info     *sbi = dp->data;
 533         struct seq_file       *seq;
 534         int                    result;
 535
 536         result = seq_open(filp, &vvp_pgcache_ops);
 537         if (result == 0) {
 538                 seq = filp->private_data;
 539                 seq->private = sbi;
 540         }
 541         return result;
 542 }
 543
 544 struct file_operations vvp_dump_pgcache_file_ops = {
 545         .owner   = THIS_MODULE,
 546         .open    = vvp_dump_pgcache_seq_open,
 547         .read    = seq_read,
 548         .llseek  = seq_lseek,
 549         .release = seq_release,
 550 };