lustre/ptlrpc/lproc_ptlrpc.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 2002 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of the Lustre file system, http://www.lustre.org
   7  *   Lustre is a trademark of Cluster File Systems, Inc.
   8  *
   9  *   You may have signed or agreed to another license before downloading
  10  *   this software.  If so, you are bound by the terms and conditions
  11  *   of that agreement, and the following does not apply to you.  See the
  12  *   LICENSE file included with this distribution for more information.
  13  *
  14  *   If you did not agree to a different license, then this copy of Lustre
  15  *   is open source software; you can redistribute it and/or modify it
  16  *   under the terms of version 2 of the GNU General Public License as
  17  *   published by the Free Software Foundation.
  18  *
  19  *   In either case, Lustre is distributed in the hope that it will be
  20  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  21  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  22  *   license text for more details.
  23  *
  24  */
  25 #define DEBUG_SUBSYSTEM S_CLASS
  26
  27 #ifndef __KERNEL__
  28 # include <liblustre.h>
  29 #endif
  30
  31 #include <obd_support.h>
  32 #include <obd.h>
  33 #include <lprocfs_status.h>
  34 #include <lustre/lustre_idl.h>
  35 #include <lustre_net.h>
  36 #include <obd_class.h>
  37 #include "ptlrpc_internal.h"
  38
  39
  40 struct ll_rpc_opcode {
  41      __u32       opcode;
  42      const char *opname;
  43 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
  44         { OST_REPLY,        "ost_reply" },
  45         { OST_GETATTR,      "ost_getattr" },
  46         { OST_SETATTR,      "ost_setattr" },
  47         { OST_READ,         "ost_read" },
  48         { OST_WRITE,        "ost_write" },
  49         { OST_CREATE ,      "ost_create" },
  50         { OST_DESTROY,      "ost_destroy" },
  51         { OST_GET_INFO,     "ost_get_info" },
  52         { OST_CONNECT,      "ost_connect" },
  53         { OST_DISCONNECT,   "ost_disconnect" },
  54         { OST_PUNCH,        "ost_punch" },
  55         { OST_OPEN,         "ost_open" },
  56         { OST_CLOSE,        "ost_close" },
  57         { OST_STATFS,       "ost_statfs" },
  58         { 14,               "ost_san_read" },
  59         { 15,               "ost_san_write" },
  60         { OST_SYNC,         "ost_sync" },
  61         { OST_SET_INFO,     "ost_set_info" },
  62         { OST_QUOTACHECK,   "ost_quotacheck" },
  63         { OST_QUOTACTL,     "ost_quotactl" },
  64         { MDS_GETATTR,      "mds_getattr" },
  65         { MDS_GETATTR_NAME, "mds_getattr_lock" },
  66         { MDS_CLOSE,        "mds_close" },
  67         { MDS_REINT,        "mds_reint" },
  68         { MDS_READPAGE,     "mds_readpage" },
  69         { MDS_CONNECT,      "mds_connect" },
  70         { MDS_DISCONNECT,   "mds_disconnect" },
  71         { MDS_GETSTATUS,    "mds_getstatus" },
  72         { MDS_STATFS,       "mds_statfs" },
  73         { MDS_PIN,          "mds_pin" },
  74         { MDS_UNPIN,        "mds_unpin" },
  75         { MDS_SYNC,         "mds_sync" },
  76         { MDS_DONE_WRITING, "mds_done_writing" },
  77         { MDS_SET_INFO,     "mds_set_info" },
  78         { MDS_QUOTACHECK,   "mds_quotacheck" },
  79         { MDS_QUOTACTL,     "mds_quotactl" },
  80         { MDS_GETXATTR,     "mds_getxattr" },
  81         { MDS_SETXATTR,     "mds_setxattr" },
  82         { MDS_WRITEPAGE,    "mds_writepage" },
  83         { MDS_IS_SUBDIR,    "mds_is_subdir" },
  84         { LDLM_ENQUEUE,     "ldlm_enqueue" },
  85         { LDLM_CONVERT,     "ldlm_convert" },
  86         { LDLM_CANCEL,      "ldlm_cancel" },
  87         { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
  88         { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
  89         { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
  90         { MGS_CONNECT,      "mgs_connect" },
  91         { MGS_DISCONNECT,   "mgs_disconnect" },
  92         { MGS_EXCEPTION,    "mgs_exception" },
  93         { MGS_TARGET_REG,   "mgs_target_reg" },
  94         { MGS_TARGET_DEL,   "mgs_target_del" },
  95         { OBD_PING,         "obd_ping" },
  96         { OBD_LOG_CANCEL,   "llog_origin_handle_cancel" },
  97         { OBD_QC_CALLBACK,  "obd_qc_callback" },
  98         { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_create" },
  99         { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
 100         { LLOG_ORIGIN_HANDLE_READ_HEADER, "llog_origin_handle_read_header" },
 101         { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
 102         { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
 103         { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
 104         { LLOG_CATINFO,     "llog_catinfo" },
 105         { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
 106         { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
 107         { FLD_QUERY,        "fld_query" },
 108         { SEQ_QUERY,        "seq_query" },
 109         { SEC_CTX_INIT,     "sec_ctx_init" },
 110         { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
 111         { SEC_CTX_FINI,     "sec_ctx_fini" }
 112 };
 113
 114 const char* ll_opcode2str(__u32 opcode)
 115 {
 116         /* When one of the assertions below fail, chances are that:
 117          *     1) A new opcode was added in lustre_idl.h, but was
 118          *        is missing from the table above.
 119          * or  2) The opcode space was renumbered or rearranged,
 120          *        and the opcode_offset() function in
 121          *        ptlrpc_internal.h needs to be modified.
 122          */
 123         __u32 offset = opcode_offset(opcode);
 124         LASSERT(offset < LUSTRE_MAX_OPCODES);
 125         LASSERT(ll_rpc_opcode_table[offset].opcode == opcode);
 126         return ll_rpc_opcode_table[offset].opname;
 127 }
 128
 129 #ifdef LPROCFS
 130 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
 131                              char *name, struct proc_dir_entry **procroot_ret,
 132                              struct lprocfs_stats **stats_ret)
 133 {
 134         struct proc_dir_entry *svc_procroot;
 135         struct lprocfs_stats *svc_stats;
 136         int i, rc;
 137         unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
 138                                           LPROCFS_CNTR_STDDEV;
 139
 140         LASSERT(*procroot_ret == NULL);
 141         LASSERT(*stats_ret == NULL);
 142
 143         svc_stats = lprocfs_alloc_stats(PTLRPC_LAST_CNTR + LUSTRE_MAX_OPCODES);
 144         if (svc_stats == NULL)
 145                 return;
 146
 147         if (dir) {
 148                 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
 149                 if (IS_ERR(svc_procroot)) {
 150                         lprocfs_free_stats(&svc_stats);
 151                         return;
 152                 }
 153         } else {
 154                 svc_procroot = root;
 155         }
 156
 157         lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
 158                              svc_counter_config, "req_waittime", "usec");
 159         lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
 160                              svc_counter_config, "req_qdepth", "reqs");
 161         lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
 162                              svc_counter_config, "req_active", "reqs");
 163         lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
 164                              svc_counter_config, "reqbuf_avail", "bufs");
 165         for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
 166                 __u32 opcode = ll_rpc_opcode_table[i].opcode;
 167                 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
 168                                      svc_counter_config, ll_opcode2str(opcode),
 169                                      "usec");
 170         }
 171
 172         rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
 173         if (rc < 0) {
 174                 if (dir)
 175                         lprocfs_remove(&svc_procroot);
 176                 lprocfs_free_stats(&svc_stats);
 177         } else {
 178                 if (dir)
 179                         *procroot_ret = svc_procroot;
 180                 *stats_ret = svc_stats;
 181         }
 182 }
 183
 184 static int
 185 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
 186                                     int count, int *eof, void *data)
 187 {
 188         struct ptlrpc_service *svc = data;
 189
 190         *eof = 1;
 191         return snprintf(page, count, "%d\n", svc->srv_n_history_rqbds);
 192 }
 193
 194 static int
 195 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
 196                                     int count, int *eof, void *data)
 197 {
 198         struct ptlrpc_service *svc = data;
 199
 200         *eof = 1;
 201         return snprintf(page, count, "%d\n", svc->srv_max_history_rqbds);
 202 }
 203
 204 static int
 205 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
 206                                      unsigned long count, void *data)
 207 {
 208         struct ptlrpc_service *svc = data;
 209         int                    bufpages;
 210         int                    val;
 211         int                    rc = lprocfs_write_helper(buffer, count, &val);
 212
 213         if (rc < 0)
 214                 return rc;
 215
 216         if (val < 0)
 217                 return -ERANGE;
 218
 219         /* This sanity check is more of an insanity check; we can still
 220          * hose a kernel by allowing the request history to grow too
 221          * far. */
 222         bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
 223         if (val > num_physpages/(2 * bufpages))
 224                 return -ERANGE;
 225
 226         spin_lock(&svc->srv_lock);
 227         svc->srv_max_history_rqbds = val;
 228         spin_unlock(&svc->srv_lock);
 229
 230         return count;
 231 }
 232
 233 struct ptlrpc_srh_iterator {
 234         __u64                  srhi_seq;
 235         struct ptlrpc_request *srhi_req;
 236 };
 237
 238 int
 239 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service *svc,
 240                                     struct ptlrpc_srh_iterator *srhi,
 241                                     __u64 seq)
 242 {
 243         struct list_head      *e;
 244         struct ptlrpc_request *req;
 245
 246         if (srhi->srhi_req != NULL &&
 247             srhi->srhi_seq > svc->srv_request_max_cull_seq &&
 248             srhi->srhi_seq <= seq) {
 249                 /* If srhi_req was set previously, hasn't been culled and
 250                  * we're searching for a seq on or after it (i.e. more
 251                  * recent), search from it onwards.
 252                  * Since the service history is LRU (i.e. culled reqs will
 253                  * be near the head), we shouldn't have to do long
 254                  * re-scans */
 255                 LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
 256                 LASSERT (!list_empty(&svc->srv_request_history));
 257                 e = &srhi->srhi_req->rq_history_list;
 258         } else {
 259                 /* search from start */
 260                 e = svc->srv_request_history.next;
 261         }
 262
 263         while (e != &svc->srv_request_history) {
 264                 req = list_entry(e, struct ptlrpc_request, rq_history_list);
 265
 266                 if (req->rq_history_seq >= seq) {
 267                         srhi->srhi_seq = req->rq_history_seq;
 268                         srhi->srhi_req = req;
 269                         return 0;
 270                 }
 271                 e = e->next;
 272         }
 273
 274         return -ENOENT;
 275 }
 276
 277 static void *
 278 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
 279 {
 280         struct ptlrpc_service       *svc = s->private;
 281         struct ptlrpc_srh_iterator  *srhi;
 282         int                          rc;
 283
 284         OBD_ALLOC(srhi, sizeof(*srhi));
 285         if (srhi == NULL)
 286                 return NULL;
 287
 288         srhi->srhi_seq = 0;
 289         srhi->srhi_req = NULL;
 290
 291         spin_lock(&svc->srv_lock);
 292         rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos);
 293         spin_unlock(&svc->srv_lock);
 294
 295         if (rc == 0) {
 296                 *pos = srhi->srhi_seq;
 297                 return srhi;
 298         }
 299
 300         OBD_FREE(srhi, sizeof(*srhi));
 301         return NULL;
 302 }
 303
 304 static void
 305 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
 306 {
 307         struct ptlrpc_srh_iterator *srhi = iter;
 308
 309         if (srhi != NULL)
 310                 OBD_FREE(srhi, sizeof(*srhi));
 311 }
 312
 313 static void *
 314 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
 315                                     void *iter, loff_t *pos)
 316 {
 317         struct ptlrpc_service       *svc = s->private;
 318         struct ptlrpc_srh_iterator  *srhi = iter;
 319         int                          rc;
 320
 321         spin_lock(&svc->srv_lock);
 322         rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos + 1);
 323         spin_unlock(&svc->srv_lock);
 324
 325         if (rc != 0) {
 326                 OBD_FREE(srhi, sizeof(*srhi));
 327                 return NULL;
 328         }
 329
 330         *pos = srhi->srhi_seq;
 331         return srhi;
 332 }
 333
 334 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
 335 {
 336         struct ptlrpc_service      *svc = s->private;
 337         struct ptlrpc_srh_iterator *srhi = iter;
 338         struct ptlrpc_request      *req;
 339         int                         rc;
 340
 341         spin_lock(&svc->srv_lock);
 342
 343         rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, srhi->srhi_seq);
 344
 345         if (rc == 0) {
 346                 req = srhi->srhi_req;
 347
 348                 /* Print common req fields.
 349                  * CAVEAT EMPTOR: we're racing with the service handler
 350                  * here.  The request could contain any old crap, so you
 351                  * must be just as careful as the service's request
 352                  * parser. Currently I only print stuff here I know is OK
 353                  * to look at coz it was set up in request_in_callback()!!! */
 354                 seq_printf(s, LPD64":%s:%s:"LPD64":%d:%s ",
 355                            req->rq_history_seq, libcfs_nid2str(req->rq_self),
 356                            libcfs_id2str(req->rq_peer), req->rq_xid,
 357                            req->rq_reqlen,ptlrpc_rqphase2str(req));
 358
 359                 if (svc->srv_request_history_print_fn == NULL)
 360                         seq_printf(s, "\n");
 361                 else
 362                         svc->srv_request_history_print_fn(s, srhi->srhi_req);
 363         }
 364
 365         spin_unlock(&svc->srv_lock);
 366
 367         return rc;
 368 }
 369
 370 static int
 371 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
 372 {
 373         static struct seq_operations sops = {
 374                 .start = ptlrpc_lprocfs_svc_req_history_start,
 375                 .stop  = ptlrpc_lprocfs_svc_req_history_stop,
 376                 .next  = ptlrpc_lprocfs_svc_req_history_next,
 377                 .show  = ptlrpc_lprocfs_svc_req_history_show,
 378         };
 379         struct proc_dir_entry *dp = PDE(inode);
 380         struct seq_file       *seqf;
 381         int                    rc;
 382
 383         LPROCFS_ENTRY_AND_CHECK(dp);
 384         rc = seq_open(file, &sops);
 385         if (rc) {
 386                 LPROCFS_EXIT();
 387                 return rc;
 388         }
 389
 390         seqf = file->private_data;
 391         seqf->private = dp->data;
 392         return 0;
 393 }
 394
 395 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
 396                                      struct ptlrpc_service *svc)
 397 {
 398         struct lprocfs_vars lproc_vars[] = {
 399                 {.name       = "req_buffer_history_len",
 400                  .write_fptr = NULL,
 401                  .read_fptr  = ptlrpc_lprocfs_read_req_history_len,
 402                  .data       = svc},
 403                 {.name       = "req_buffer_history_max",
 404                  .write_fptr = ptlrpc_lprocfs_write_req_history_max,
 405                  .read_fptr  = ptlrpc_lprocfs_read_req_history_max,
 406                  .data       = svc},
 407                 {NULL}
 408         };
 409         static struct file_operations req_history_fops = {
 410                 .owner       = THIS_MODULE,
 411                 .open        = ptlrpc_lprocfs_svc_req_history_open,
 412                 .read        = seq_read,
 413                 .llseek      = seq_lseek,
 414                 .release     = lprocfs_seq_release,
 415         };
 416         struct proc_dir_entry *req_history;
 417
 418         ptlrpc_lprocfs_register(entry, svc->srv_name,
 419                                 "stats", &svc->srv_procroot,
 420                                 &svc->srv_stats);
 421
 422         if (svc->srv_procroot == NULL)
 423                 return;
 424
 425         lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
 426
 427         req_history = create_proc_entry("req_history", 0400,
 428                                         svc->srv_procroot);
 429         if (req_history != NULL) {
 430                 req_history->data = svc;
 431                 req_history->proc_fops = &req_history_fops;
 432         }
 433 }
 434
 435 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
 436 {
 437         ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
 438                                 &obddev->obd_svc_procroot,
 439                                 &obddev->obd_svc_stats);
 440 }
 441 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
 442
 443 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req)
 444 {
 445         struct lprocfs_stats *svc_stats;
 446         int opc = opcode_offset(lustre_msg_get_opc(req->rq_reqmsg));
 447
 448         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 449         if (svc_stats == NULL || opc <= 0)
 450                 return;
 451         LASSERT(opc < LUSTRE_MAX_OPCODES);
 452         /* These two use the ptlrpc_lprocfs_brw below */
 453         if (!(opc == OST_WRITE || opc == OST_READ))
 454                 lprocfs_counter_add(svc_stats, opc + PTLRPC_LAST_CNTR, 0);
 455 }
 456
 457 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes)
 458 {
 459         struct lprocfs_stats *svc_stats;
 460         svc_stats = req->rq_import->imp_obd->obd_svc_stats;
 461         if (!svc_stats)
 462                 return;
 463         lprocfs_counter_add(svc_stats, opc + PTLRPC_LAST_CNTR, bytes);
 464 }
 465 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
 466
 467 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
 468 {
 469         if (svc->srv_procroot != NULL)
 470                 lprocfs_remove(&svc->srv_procroot);
 471
 472         if (svc->srv_stats)
 473                 lprocfs_free_stats(&svc->srv_stats);
 474 }
 475
 476 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
 477 {
 478         if (obd->obd_svc_procroot)
 479                 lprocfs_remove(&obd->obd_svc_procroot);
 480
 481         if (obd->obd_svc_stats)
 482                 lprocfs_free_stats(&obd->obd_svc_stats);
 483 }
 484 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
 485
 486
 487 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
 488                             unsigned long count, void *data)
 489 {
 490         struct obd_device *obd = data;
 491         char tmpbuf[sizeof(struct obd_uuid)];
 492
 493         /* Kludge code(deadlock situation): the lprocfs lock has been held
 494          * since the client is evicted by writting client's
 495          * uuid/nid to procfs "evict_client" entry. However,
 496          * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
 497          * the proc entries under the being destroyed export{}, so I have
 498          * to drop the lock at first here.
 499          * - jay, jxiong@clusterfs.com */
 500         class_incref(obd);
 501         LPROCFS_EXIT();
 502
 503         sscanf(buffer, "%40s", tmpbuf);
 504         if (strncmp(tmpbuf, "nid:", 4) == 0)
 505                 obd_export_evict_by_nid(obd, tmpbuf + 4);
 506         else if (strncmp(tmpbuf, "uuid:", 5) == 0)
 507                 obd_export_evict_by_uuid(obd, tmpbuf + 5);
 508         else
 509                 obd_export_evict_by_uuid(obd, tmpbuf);
 510
 511         LPROCFS_ENTRY();
 512         class_decref(obd);
 513
 514         return count;
 515 }
 516 EXPORT_SYMBOL(lprocfs_wr_evict_client);
 517
 518 int lprocfs_wr_ping(struct file *file, const char *buffer,
 519                     unsigned long count, void *data)
 520 {
 521         struct obd_device *obd = data;
 522         struct ptlrpc_request *req;
 523         int rc;
 524         ENTRY;
 525
 526         LPROCFS_CLIMP_CHECK(obd);
 527         req = ptlrpc_prep_req(obd->u.cli.cl_import, LUSTRE_OBD_VERSION,
 528                               OBD_PING, 1, NULL, NULL);
 529         LPROCFS_CLIMP_EXIT(obd);
 530         if (req == NULL)
 531                 RETURN(-ENOMEM);
 532
 533         ptlrpc_req_set_repsize(req, 1, NULL);
 534         req->rq_send_state = LUSTRE_IMP_FULL;
 535         req->rq_no_resend = 1;
 536
 537         rc = ptlrpc_queue_wait(req);
 538
 539         ptlrpc_req_finished(req);
 540         if (rc >= 0)
 541                 RETURN(count);
 542         RETURN(rc);
 543 }
 544 EXPORT_SYMBOL(lprocfs_wr_ping);
 545
 546 #endif /* LPROCFS */