multipath-tools/patches/0024-RHBZ-565933-checker-timeout.patch

   1 ---
   2  libmultipath/checkers.h              |   15 +--------------
   3  libmultipath/checkers/emc_clariion.c |    4 ++--
   4  libmultipath/checkers/hp_sw.c        |   12 ++++++------
   5  libmultipath/checkers/libsg.c        |    5 +++--
   6  libmultipath/checkers/libsg.h        |    3 ++-
   7  libmultipath/checkers/rdac.c         |    9 +++++----
   8  libmultipath/checkers/readsector0.c  |    2 +-
   9  libmultipath/checkers/tur.c          |    4 ++--
  10  libmultipath/config.h                |    1 +
  11  libmultipath/dict.c                  |   29 +++++++++++++++++++++++++++++
  12  libmultipath/discovery.c             |   27 +++++++++++++++++++++++++++
  13  libmultipath/discovery.h             |    1 +
  14  libmultipath/propsel.c               |   19 +++++++++++++++++--
  15  multipath.conf.annotated             |    9 +++++++++
  16  14 files changed, 106 insertions(+), 34 deletions(-)
  17
  18 Index: multipath-tools/libmultipath/checkers.h
  19 ===================================================================
  20 --- multipath-tools.orig/libmultipath/checkers.h
  21 +++ multipath-tools/libmultipath/checkers.h
  22 @@ -69,20 +69,6 @@ enum path_check_state {
  23
  24  #define DEFAULT_CHECKER DIRECTIO
  25
  26 -/*
  27 - * Overloaded storage response time can be very long.
  28 - * SG_IO timouts after DEF_TIMEOUT milliseconds, and checkers interprets this
  29 - * as a path failure. multipathd then proactively evicts the path from the DM
  30 - * multipath table in this case.
  31 - *
  32 - * This generaly snow balls and ends up in full eviction and IO errors for end
  33 - * users. Bad. This may also cause SCSI bus resets, causing disruption for all
  34 - * local and external storage hardware users.
  35 - *
  36 - * Provision a long timeout. Longer than any real-world application would cope
  37 - * with.
  38 - */
  39 -#define DEF_TIMEOUT            300000
  40  #define ASYNC_TIMEOUT_SEC      30
  41
  42  /*
  43 @@ -98,6 +84,7 @@ struct checker {
  44         struct list_head node;
  45         int fd;
  46         int sync;
  47 +       unsigned int timeout;
  48         int disable;
  49         char name[CHECKER_NAME_LEN];
  50         char message[CHECKER_MSG_LEN];       /* comm with callers */
  51 Index: multipath-tools/libmultipath/checkers/emc_clariion.c
  52 ===================================================================
  53 --- multipath-tools.orig/libmultipath/checkers/emc_clariion.c
  54 +++ multipath-tools/libmultipath/checkers/emc_clariion.c
  55 @@ -113,7 +113,7 @@ int libcheck_check (struct checker * c)
  56         io_hdr.dxferp = sense_buffer;
  57         io_hdr.cmdp = inqCmdBlk;
  58         io_hdr.sbp = sb;
  59 -       io_hdr.timeout = DEF_TIMEOUT;
  60 +       io_hdr.timeout = c->timeout;
  61         io_hdr.pack_id = 0;
  62         if (ioctl(c->fd, SG_IO, &io_hdr) < 0) {
  63                 MSG(c, "emc_clariion_checker: sending query command failed");
  64 @@ -182,7 +182,7 @@ int libcheck_check (struct checker * c)
  65                 unsigned char buf[4096];
  66
  67                 memset(buf, 0, 4096);
  68 -               ret = sg_read(c->fd, &buf[0], sbb = &sb[0]);
  69 +               ret = sg_read(c->fd, &buf[0], sbb = &sb[0], c->timeout);
  70                 if (ret == PATH_DOWN) {
  71                         hexadecimal_to_ascii(ct->wwn, wwnstr);
  72
  73 Index: multipath-tools/libmultipath/checkers/hp_sw.c
  74 ===================================================================
  75 --- multipath-tools.orig/libmultipath/checkers/hp_sw.c
  76 +++ multipath-tools/libmultipath/checkers/hp_sw.c
  77 @@ -46,7 +46,7 @@ void libcheck_free (struct checker * c)
  78
  79  static int
  80  do_inq(int sg_fd, int cmddt, int evpd, unsigned int pg_op,
  81 -       void *resp, int mx_resp_len, int noisy)
  82 +       void *resp, int mx_resp_len, int noisy, unsigned int timeout)
  83  {
  84         unsigned char inqCmdBlk[INQUIRY_CMDLEN] =
  85                 { INQUIRY_CMD, 0, 0, 0, 0, 0 };
  86 @@ -70,7 +70,7 @@ do_inq(int sg_fd, int cmddt, int evpd, u
  87         io_hdr.dxferp = resp;
  88         io_hdr.cmdp = inqCmdBlk;
  89         io_hdr.sbp = sense_b;
  90 -       io_hdr.timeout = DEF_TIMEOUT;
  91 +       io_hdr.timeout = timeout;
  92
  93         if (ioctl(sg_fd, SG_IO, &io_hdr) < 0)
  94                 return 1;
  95 @@ -98,7 +98,7 @@ do_inq(int sg_fd, int cmddt, int evpd, u
  96  }
  97
  98  static int
  99 -do_tur (int fd)
 100 +do_tur (int fd, unsigned int timeout)
 101  {
 102         unsigned char turCmdBlk[TUR_CMD_LEN] = { 0x00, 0, 0, 0, 0, 0 };
 103         struct sg_io_hdr io_hdr;
 104 @@ -111,7 +111,7 @@ do_tur (int fd)
 105         io_hdr.dxfer_direction = SG_DXFER_NONE;
 106         io_hdr.cmdp = turCmdBlk;
 107         io_hdr.sbp = sense_buffer;
 108 -       io_hdr.timeout = DEF_TIMEOUT;
 109 +       io_hdr.timeout = timeout;
 110         io_hdr.pack_id = 0;
 111
 112         if (ioctl(fd, SG_IO, &io_hdr) < 0)
 113 @@ -128,12 +128,12 @@ libcheck_check (struct checker * c)
 114  {
 115         char buff[MX_ALLOC_LEN];
 116
 117 -       if (0 != do_inq(c->fd, 0, 1, 0x80, buff, MX_ALLOC_LEN, 0)) {
 118 +       if (0 != do_inq(c->fd, 0, 1, 0x80, buff, MX_ALLOC_LEN, 0, c->timeout)) {
 119                 MSG(c, MSG_HP_SW_DOWN);
 120                 return PATH_DOWN;
 121         }
 122
 123 -       if (do_tur(c->fd)) {
 124 +       if (do_tur(c->fd, c->timeout)) {
 125                 MSG(c, MSG_HP_SW_GHOST);
 126                 return PATH_GHOST;
 127         }
 128 Index: multipath-tools/libmultipath/checkers/libsg.c
 129 ===================================================================
 130 --- multipath-tools.orig/libmultipath/checkers/libsg.c
 131 +++ multipath-tools/libmultipath/checkers/libsg.c
 132 @@ -11,7 +11,8 @@
 133  #include "../libmultipath/sg_include.h"
 134
 135  int
 136 -sg_read (int sg_fd, unsigned char * buff, unsigned char * senseBuff)
 137 +sg_read (int sg_fd, unsigned char * buff, unsigned char * senseBuff,
 138 +        unsigned int timeout)
 139  {
 140         /* defaults */
 141         int blocks = 1;
 142 @@ -51,7 +52,7 @@ sg_read (int sg_fd, unsigned char * buff
 143         io_hdr.dxferp = buff;
 144         io_hdr.mx_sb_len = SENSE_BUFF_LEN;
 145         io_hdr.sbp = senseBuff;
 146 -       io_hdr.timeout = DEF_TIMEOUT;
 147 +       io_hdr.timeout = timeout;
 148         io_hdr.pack_id = (int)start_block;
 149         if (diop && *diop)
 150         io_hdr.flags |= SG_FLAG_DIRECT_IO;
 151 Index: multipath-tools/libmultipath/checkers/libsg.h
 152 ===================================================================
 153 --- multipath-tools.orig/libmultipath/checkers/libsg.h
 154 +++ multipath-tools/libmultipath/checkers/libsg.h
 155 @@ -3,6 +3,7 @@
 156
 157  #define SENSE_BUFF_LEN 32
 158
 159 -int sg_read (int sg_fd, unsigned char * buff, unsigned char * senseBuff);
 160 +int sg_read (int sg_fd, unsigned char * buff, unsigned char * senseBuff,
 161 +            unsigned int timeout);
 162
 163  #endif /* _LIBSG_H */
 164 Index: multipath-tools/libmultipath/checkers/rdac.c
 165 ===================================================================
 166 --- multipath-tools.orig/libmultipath/checkers/rdac.c
 167 +++ multipath-tools/libmultipath/checkers/rdac.c
 168 @@ -18,7 +18,6 @@
 169  #define INQUIRY_CMDLEN         6
 170  #define INQUIRY_CMD            0x12
 171  #define SENSE_BUFF_LEN         32
 172 -#define RDAC_DEF_TIMEOUT       60000
 173  #define SCSI_CHECK_CONDITION   0x2
 174  #define SCSI_COMMAND_TERMINATED        0x22
 175  #define SG_ERR_DRIVER_SENSE    0x08
 176 @@ -43,7 +42,8 @@ void libcheck_free (struct checker * c)
 177  }
 178
 179  static int
 180 -do_inq(int sg_fd, unsigned int pg_op, void *resp, int mx_resp_len)
 181 +do_inq(int sg_fd, unsigned int pg_op, void *resp, int mx_resp_len,
 182 +       unsigned int timeout)
 183  {
 184         unsigned char inqCmdBlk[INQUIRY_CMDLEN] = { INQUIRY_CMD, 1, 0, 0, 0, 0 };
 185         unsigned char sense_b[SENSE_BUFF_LEN];
 186 @@ -62,7 +62,7 @@ do_inq(int sg_fd, unsigned int pg_op, vo
 187         io_hdr.dxferp = resp;
 188         io_hdr.cmdp = inqCmdBlk;
 189         io_hdr.sbp = sense_b;
 190 -       io_hdr.timeout = RDAC_DEF_TIMEOUT;
 191 +       io_hdr.timeout = timeout;
 192
 193         if (ioctl(sg_fd, SG_IO, &io_hdr) < 0)
 194                 return 1;
 195 @@ -103,7 +103,8 @@ libcheck_check (struct checker * c)
 196         struct volume_access_inq inq;
 197
 198         memset(&inq, 0, sizeof(struct volume_access_inq));
 199 -       if (0 != do_inq(c->fd, 0xC9, &inq, sizeof(struct volume_access_inq))) {
 200 +       if (0 != do_inq(c->fd, 0xC9, &inq, sizeof(struct volume_access_inq),
 201 +                       c->timeout)) {
 202                 MSG(c, MSG_RDAC_DOWN);
 203                 return PATH_DOWN;
 204         } else {
 205 Index: multipath-tools/libmultipath/checkers/readsector0.c
 206 ===================================================================
 207 --- multipath-tools.orig/libmultipath/checkers/readsector0.c
 208 +++ multipath-tools/libmultipath/checkers/readsector0.c
 209 @@ -29,7 +29,7 @@ int libcheck_check (struct checker * c)
 210         unsigned char sbuf[SENSE_BUFF_LEN];
 211         int ret;
 212
 213 -       ret = sg_read(c->fd, &buf[0], &sbuf[0]);
 214 +       ret = sg_read(c->fd, &buf[0], &sbuf[0], c->timeout);
 215
 216         switch (ret)
 217         {
 218 Index: multipath-tools/libmultipath/checkers/tur.c
 219 ===================================================================
 220 --- multipath-tools.orig/libmultipath/checkers/tur.c
 221 +++ multipath-tools/libmultipath/checkers/tur.c
 222 @@ -63,7 +63,7 @@ retry:
 223         io_hdr.dxferp = (unsigned char *)resp_buffer;
 224         io_hdr.cmdp = inq_cmd;
 225         io_hdr.sbp = sense_buffer;
 226 -       io_hdr.timeout = 60; // IOCTL timeout value.
 227 +       io_hdr.timeout = c->timeout; // IOCTL timeout value.
 228
 229         if (ioctl(c->fd, SG_IO, &io_hdr) < 0) {
 230                 condlog(0, "SG_IO ioctl failed: %s", strerror(errno));
 231 @@ -148,7 +148,7 @@ libcheck_check (struct checker * c)
 232         io_hdr.dxfer_direction = SG_DXFER_NONE;
 233         io_hdr.cmdp = turCmdBlk;
 234         io_hdr.sbp = sense_buffer;
 235 -       io_hdr.timeout = DEF_TIMEOUT;
 236 +       io_hdr.timeout = c->timeout;
 237         io_hdr.pack_id = 0;
 238         if (ioctl(c->fd, SG_IO, &io_hdr) < 0) {
 239                 MSG(c, MSG_TUR_DOWN);
 240 Index: multipath-tools/libmultipath/config.h
 241 ===================================================================
 242 --- multipath-tools.orig/libmultipath/config.h
 243 +++ multipath-tools/libmultipath/config.h
 244 @@ -80,6 +80,7 @@ struct config {
 245         int max_fds;
 246         int force_reload;
 247         int queue_without_daemon;
 248 +       int checker_timeout;
 249         int daemon;
 250         int flush_on_last_del;
 251         int attribute_flags;
 252 Index: multipath-tools/libmultipath/dict.c
 253 ===================================================================
 254 --- multipath-tools.orig/libmultipath/dict.c
 255 +++ multipath-tools/libmultipath/dict.c
 256 @@ -396,6 +396,25 @@ def_queue_without_daemon(vector strvec)
 257  }
 258
 259  static int
 260 +def_checker_timeout_handler(vector strvec)
 261 +{
 262 +       unsigned int checker_timeout;
 263 +       char *buff;
 264 +
 265 +       buff = set_value(strvec);
 266 +       if (!buff)
 267 +               return 1;
 268 +
 269 +       if (sscanf(buff, "%u", &checker_timeout) == 1)
 270 +               conf->checker_timeout = checker_timeout;
 271 +       else
 272 +               conf->checker_timeout = 0;
 273 +
 274 +       free(buff);
 275 +       return 0;
 276 +}
 277 +
 278 +static int
 279  def_pg_timeout_handler(vector strvec)
 280  {
 281         int pg_timeout;
 282 @@ -2068,6 +2087,15 @@ snprint_def_queue_without_daemon (char *
 283  }
 284
 285  static int
 286 +snprint_def_checker_timeout (char *buff, int len, void *data)
 287 +{
 288 +       if (!conf->checker_timeout)
 289 +               return 0;
 290 +
 291 +       return snprintf(buff, len, "%u", conf->checker_timeout);
 292 +}
 293 +
 294 +static int
 295  snprint_def_pg_timeout (char * buff, int len, void * data)
 296  {
 297         if (conf->pg_timeout == DEFAULT_PGTIMEOUT)
 298 @@ -2166,6 +2194,7 @@ init_keywords(void)
 299         install_keyword("rr_weight", &def_weight_handler, &snprint_def_rr_weight);
 300         install_keyword("no_path_retry", &def_no_path_retry_handler, &snprint_def_no_path_retry);
 301         install_keyword("queue_without_daemon", &def_queue_without_daemon, &snprint_def_queue_without_daemon);
 302 +       install_keyword("checker_timeout", &def_checker_timeout_handler, &snprint_def_checker_timeout);
 303         install_keyword("pg_timeout", &def_pg_timeout_handler, &snprint_def_pg_timeout);
 304         install_keyword("flush_on_last_del", &def_flush_on_last_del_handler, &snprint_def_flush_on_last_del);
 305         install_keyword("user_friendly_names", &names_handler, &snprint_def_user_friendly_names);
 306 Index: multipath-tools/libmultipath/discovery.c
 307 ===================================================================
 308 --- multipath-tools.orig/libmultipath/discovery.c
 309 +++ multipath-tools/libmultipath/discovery.c
 310 @@ -164,6 +164,31 @@ sysfs_get_dev (struct sysfs_device * dev
 311  }
 312
 313  int
 314 +sysfs_get_timeout(struct sysfs_device *dev, unsigned int *timeout)
 315 +{
 316 +       char *attr;
 317 +       char attr_path[SYSFS_PATH_SIZE];
 318 +       int r;
 319 +       unsigned int t;
 320 +
 321 +       if (safe_sprintf(attr_path, "%s/device", dev->devpath))
 322 +               return 1;
 323 +
 324 +       attr = sysfs_attr_get_value(dev->devpath, "timeout");
 325 +       if (!attr)
 326 +               return 1;
 327 +
 328 +       r = sscanf(attr, "%u\n", &t);
 329 +
 330 +       if (r != 1)
 331 +               return 1;
 332 +
 333 +       *timeout = t * 1000;
 334 +
 335 +       return 0;
 336 +}
 337 +
 338 +int
 339  sysfs_get_size (struct sysfs_device * dev, unsigned long long * size)
 340  {
 341         char *attr;
 342 @@ -791,6 +816,8 @@ get_state (struct path * pp, int daemon)
 343                         return PATH_PENDING;
 344                 checker_set_async(c);
 345         }
 346 +       if (!conf->checker_timeout)
 347 +               sysfs_get_timeout(pp->sysdev, &(c->timeout));
 348         state = checker_check(c);
 349         condlog(3, "%s: state = %i", pp->dev, state);
 350         if (state == PATH_DOWN && strlen(checker_message(c)))
 351 Index: multipath-tools/libmultipath/propsel.c
 352 ===================================================================
 353 --- multipath-tools.orig/libmultipath/propsel.c
 354 +++ multipath-tools/libmultipath/propsel.c
 355 @@ -16,6 +16,7 @@
 356  #include "defaults.h"
 357  #include "devmapper.h"
 358  #include "prio.h"
 359 +#include "discovery.h"
 360
 361  pgpolicyfn *pgpolicies[] = {
 362         NULL,
 363 @@ -274,17 +275,31 @@ select_checker(struct path *pp)
 364                 checker_get(c, pp->hwe->checker_name);
 365                 condlog(3, "%s: path checker = %s (controller setting)",
 366                         pp->dev, checker_name(c));
 367 -               return 0;
 368 +               goto out;
 369         }
 370         if (conf->checker_name) {
 371                 checker_get(c, conf->checker_name);
 372                 condlog(3, "%s: path checker = %s (config file default)",
 373                         pp->dev, checker_name(c));
 374 -               return 0;
 375 +               goto out;
 376         }
 377         checker_get(c, DEFAULT_CHECKER);
 378         condlog(3, "%s: path checker = %s (internal default)",
 379                 pp->dev, checker_name(c));
 380 +out:
 381 +       if (conf->checker_timeout) {
 382 +               c->timeout = conf->checker_timeout * 1000;
 383 +               condlog(3, "%s: checker timeout = %u ms (config file default)",
 384 +                               pp->dev, c->timeout);
 385 +       }
 386 +       else if (sysfs_get_timeout(pp->sysdev, &c->timeout) == 0)
 387 +               condlog(3, "%s: checker timeout = %u ms (sysfs setting)",
 388 +                               pp->dev, c->timeout);
 389 +       else {
 390 +               c->timeout = DEF_TIMEOUT;
 391 +               condlog(3, "%s: checker timeout = %u ms (internal default)",
 392 +                               pp->dev, c->timeout);
 393 +       }
 394         return 0;
 395  }
 396
 397 Index: multipath-tools/multipath.conf.annotated
 398 ===================================================================
 399 --- multipath-tools.orig/multipath.conf.annotated
 400 +++ multipath-tools/multipath.conf.annotated
 401 @@ -202,6 +202,15 @@
 402  #      gid disk
 403  #
 404  #      #
 405 +#      # name    : checker_timeout
 406 +#      # scope   : multipath & multipathd
 407 +#      # desc    : The timeout to use for path checkers that issue scsi
 408 +#      #           commands with an explicit timeout, in seconds.
 409 +#      # values  : n > 0
 410 +#      # default : taken from /sys/block/sd<x>/device/timeout
 411 +#      checker_timeout 60
 412 +#
 413 +#      #
 414  #      # name    : fast_io_fail_tmo
 415  #      # scope   : multipath & multipathd
 416  #      # desc    : The number of seconds the scsi layer will wait after a
 417 Index: multipath-tools/libmultipath/discovery.h
 418 ===================================================================
 419 --- multipath-tools.orig/libmultipath/discovery.h
 420 +++ multipath-tools/libmultipath/discovery.h
 421 @@ -36,6 +36,7 @@ int pathinfo (struct path *, vector hwta
 422  struct path * store_pathinfo (vector pathvec, vector hwtable,
 423                               char * devname, int flag);
 424  int sysfs_set_scsi_tmo (struct multipath *mpp);
 425 +int sysfs_get_timeout(struct sysfs_device *dev, unsigned int *timeout);
 426
 427  /*
 428   * discovery bitmask