#ifndef _OBD_SUPPORT
#define _OBD_SUPPORT
-#include <libcfs/kp30.h>
+#include <libcfs/libcfs.h>
+#include <lvfs.h>
+#include <lprocfs_status.h>
+
+#if defined(__linux__)
+#include <linux/obd_support.h>
+#elif defined(__APPLE__)
+#include <darwin/obd_support.h>
+#elif defined(__WINNT__)
+#include <winnt/obd_support.h>
+#else
+#error Unsupported operating system.
+#endif
/* global variables */
-extern atomic_t obd_memory;
-extern int obd_memmax;
-extern unsigned int obd_fail_loc;
+extern struct lprocfs_stats *obd_memory;
+enum {
+ OBD_MEMORY_STAT = 0,
+ OBD_MEMORY_PAGES_STAT = 1,
+ OBD_STATS_NUM,
+};
+
+enum {
+ OBD_FAIL_LOC_NOSET = 0,
+ OBD_FAIL_LOC_ORSET = 1,
+ OBD_FAIL_LOC_RESET = 2
+};
+
+extern unsigned long obd_fail_loc;
+extern unsigned int obd_fail_val;
extern unsigned int obd_debug_peer_on_timeout;
extern unsigned int obd_dump_on_timeout;
extern unsigned int obd_dump_on_eviction;
extern atomic_t obd_dirty_pages;
extern cfs_waitq_t obd_race_waitq;
extern int obd_race_state;
+extern unsigned int obd_alloc_fail_rate;
+
+int __obd_fail_check_set(__u32 id, __u32 value, int set);
+
+/* lvfs.c */
+int obd_alloc_fail(const void *ptr, const char *name, const char *type,
+ size_t size, const char *file, int line);
/* Timeout definitions */
#define LDLM_TIMEOUT_DEFAULT 20
#define OBD_FAIL_MDS_WRITEPAGE_NET 0x13a
#define OBD_FAIL_MDS_WRITEPAGE_PACK 0x13b
#define OBD_FAIL_MDS_LLOG_CREATE_FAILED 0x13c
+#define OBD_FAIL_MDS_OSC_PRECREATE 0x13d
+#define OBD_FAIL_MDS_LOV_SYNC_RACE 0x13e
+#define OBD_FAIL_MDS_CLOSE_NET_REP 0x13f
+#define OBD_FAIL_MDS_LLOG_SYNC_TIMEOUT 0x140
#define OBD_FAIL_OST 0x200
#define OBD_FAIL_OST_CONNECT_NET 0x201
#define OBD_FAIL_OST_DROP_REQ 0x21d
#define OBD_FAIL_OST_SETATTR_CREDITS 0x21e
#define OBD_FAIL_OST_HOLD_WRITE_RPC 0x21f
+#define OBD_FAIL_OST_BRW_WRITE_BULK2 0x220
+#define OBD_FAIL_OST_LLOG_RECOVERY_TIMEOUT 0x221
+#define OBD_FAIL_OST_CANCEL_COOKIE_TIMEOUT 0x222
+#define OBD_FAIL_OST_CONNECT_NET2 0x225
#define OBD_FAIL_LDLM 0x300
#define OBD_FAIL_LDLM_NAMESPACE_NEW 0x301
#define OBD_FAIL_LDLM_RECOV_CLIENTS 0x30d
#define OBD_FAIL_LDLM_ENQUEUE_OLD_EXPORT 0x30e
#define OBD_FAIL_LDLM_GLIMPSE 0x30f
+#define OBD_FAIL_LDLM_CANCEL_RACE 0x310
+#define OBD_FAIL_LDLM_CANCEL_EVICT_RACE 0x311
+/*
+#define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
+*/
+#define OBD_FAIL_LDLM_CLOSE_THREAD 0x313
+#define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314
#define OBD_FAIL_OSC 0x400
#define OBD_FAIL_OSC_BRW_READ_BULK 0x401
#define OBD_FAIL_OSC_SHUTDOWN 0x407
#define OBD_FAIL_OSC_CHECKSUM_RECEIVE 0x408
#define OBD_FAIL_OSC_CHECKSUM_SEND 0x409
+#define OBD_FAIL_OSC_BRW_PREP_REQ2 0x40a
+#define OBD_FAIL_OSC_CONNECT_CKSUM 0x40b
+#define OBD_FAIL_OSC_CKSUM_ADLER_ONLY 0x40c
+#define OBD_FAIL_OSC_DIO_PAUSE 0x40d
#define OBD_FAIL_PTLRPC 0x500
#define OBD_FAIL_PTLRPC_ACK 0x501
#define OBD_FAIL_TGT_DELAY_CONNECT 0x703
#define OBD_FAIL_TGT_DELAY_RECONNECT 0x704
#define OBD_FAIL_TGT_DELAY_PRECREATE 0x705
+#define OBD_FAIL_TGT_TOOMANY_THREADS 0x706
#define OBD_FAIL_MDC_REVALIDATE_PAUSE 0x800
#define OBD_FAIL_MDC_ENQUEUE_PAUSE 0x801
+#define OBD_FAIL_MDC_GETATTR_ENQUEUE 0x803
#define OBD_FAIL_MGS 0x900
#define OBD_FAIL_MGS_ALL_REQUEST_NET 0x901
#define OBD_FAIL_LPROC_REMOVE 0xb00
+#define OBD_FAIL_GENERAL_ALLOC 0xc00
+
#define OBD_FAIL_SEQ 0x1000
#define OBD_FAIL_SEQ_QUERY_NET 0x1001
#define OBD_FAIL_SEC_CTX_INIT_CONT_NET 0x1202
#define OBD_FAIL_SEC_CTX_FINI_NET 0x1203
-/* preparation for a more advanced failure testbed (not functional yet) */
+/* Failure injection control */
#define OBD_FAIL_MASK_SYS 0x0000FF00
-#define OBD_FAIL_MASK_LOC (0x000000FF | OBD_FAIL_MASK_SYS)
-#define OBD_FAIL_ONCE 0x80000000
-#define OBD_FAILED 0x40000000
-
-#define OBD_FAIL_CHECK(id) (((obd_fail_loc & OBD_FAIL_MASK_LOC) == \
- ((id) & OBD_FAIL_MASK_LOC)) && \
- ((obd_fail_loc & (OBD_FAILED | OBD_FAIL_ONCE))!= \
- (OBD_FAILED | OBD_FAIL_ONCE)))
-
-#define OBD_FAIL_CHECK_ONCE(id) \
-({ int _ret_ = 0; \
- if (unlikely(OBD_FAIL_CHECK(id))) { \
- CERROR("*** obd_fail_loc=0x%x ***\n", id); \
- obd_fail_loc |= OBD_FAILED; \
- if ((id) & OBD_FAIL_ONCE) \
- obd_fail_loc |= OBD_FAIL_ONCE; \
- _ret_ = 1; \
- } \
- _ret_; \
-})
+#define OBD_FAIL_MASK_LOC (0x000000FF | OBD_FAIL_MASK_SYS)
-#define OBD_FAIL_RETURN(id, ret) \
-do { \
- if (unlikely(OBD_FAIL_CHECK_ONCE(id))) { \
- RETURN(ret); \
- } \
-} while(0)
+#define OBD_FAILED_BIT 30
+/* OBD_FAILED is 0x40000000 */
+#define OBD_FAILED (1 << OBD_FAILED_BIT)
-#define OBD_FAIL_TIMEOUT(id, secs) \
-do { \
- if (unlikely(OBD_FAIL_CHECK_ONCE(id))) { \
- CERROR("obd_fail_timeout id %x sleeping for %d secs\n", \
- (id), (secs)); \
- set_current_state(TASK_UNINTERRUPTIBLE); \
- cfs_schedule_timeout(CFS_TASK_UNINT, \
- cfs_time_seconds(secs)); \
- set_current_state(TASK_RUNNING); \
- CERROR("obd_fail_timeout id %x awake\n", (id)); \
- } \
-} while(0)
+#define OBD_FAIL_ONCE_BIT 31
+/* OBD_FAIL_ONCE is 0x80000000 */
+#define OBD_FAIL_ONCE (1 << OBD_FAIL_ONCE_BIT)
+
+/* The following flags aren't made to be combined */
+#define OBD_FAIL_SKIP 0x20000000 /* skip N times then fail */
+#define OBD_FAIL_SOME 0x10000000 /* only fail N times */
+#define OBD_FAIL_RAND 0x08000000 /* fail 1/N of the times */
+#define OBD_FAIL_USR1 0x04000000 /* user flag */
+
+#define OBD_FAIL_PRECHECK(id) (obd_fail_loc && \
+ (obd_fail_loc & OBD_FAIL_MASK_LOC) == \
+ ((id) & OBD_FAIL_MASK_LOC))
+
+static inline int obd_fail_check_set(__u32 id, __u32 value, int set)
+{
+ int ret = 0;
+ if (unlikely(OBD_FAIL_PRECHECK(id) &&
+ (ret = __obd_fail_check_set(id, value, set)))) {
+ CERROR("*** obd_fail_loc=%x ***\n", id);
+ }
+ return ret;
+}
+
+/* If id hit obd_fail_loc, return 1, otherwise return 0 */
+#define OBD_FAIL_CHECK(id) \
+ obd_fail_check_set(id, 0, OBD_FAIL_LOC_NOSET)
+
+/* If id hit obd_fail_loc, obd_fail_loc |= value and return 1,
+ * otherwise return 0 */
+#define OBD_FAIL_CHECK_ORSET(id, value) \
+ obd_fail_check_set(id, value, OBD_FAIL_LOC_ORSET)
+
+/* If id hit obd_fail_loc, obd_fail_loc = value and return 1,
+ * otherwise return 0 */
+#define OBD_FAIL_CHECK_RESET(id, value) \
+ obd_fail_check_set(id, value, OBD_FAIL_LOC_RESET)
+
+
+static inline int obd_fail_timeout_set(__u32 id, __u32 value, int secs, int set)
+{
+ int ret = 0;
+ if (unlikely(OBD_FAIL_PRECHECK(id) &&
+ (ret = __obd_fail_check_set(id, value, set)))) {
+ CERROR("obd_fail_timeout id %x sleeping for %d secs\n",
+ id, secs);
+ set_current_state(TASK_UNINTERRUPTIBLE);
+ cfs_schedule_timeout(CFS_TASK_UNINT, cfs_time_seconds(secs));
+ set_current_state(TASK_RUNNING);
+ CERROR("obd_fail_timeout id %x awake\n", id);
+ }
+ return ret;
+}
+
+/* If id hit obd_fail_loc, sleep secs */
+#define OBD_FAIL_TIMEOUT(id, secs) \
+ obd_fail_timeout_set(id, 0, secs, OBD_FAIL_LOC_NOSET)
+
+/* If id hit obd_fail_loc, obd_fail_loc |= value and sleep secs */
+#define OBD_FAIL_TIMEOUT_ORSET(id, value, secs) \
+ obd_fail_timeout_set(id, value, secs, OBD_FAIL_LOC_ORSET)
#ifdef __KERNEL__
+static inline void obd_fail_write(int id, struct super_block *sb)
+{
+ /* We set FAIL_ONCE because we never "un-fail" a device */
+ if (OBD_FAIL_CHECK_ORSET(id & ~OBD_FAIL_ONCE, OBD_FAIL_ONCE)) {
+#ifdef LIBCFS_DEBUG
+ BDEVNAME_DECLARE_STORAGE(tmp);
+ CERROR("obd_fail_loc=%x, fail write operation on %s\n",
+ id, ll_bdevname(sb, tmp));
+#endif
+ /* TODO-CMD: fix getting jdev */
+ __lvfs_set_rdonly(lvfs_sbdev(sb), (lvfs_sbdev_type)0);
+ }
+}
+#define OBD_FAIL_WRITE(id, sb) obd_fail_write(id, sb)
+
/* The idea here is to synchronise two threads to force a race. The
* first thread that calls this with a matching fail_loc is put to
* sleep. The next thread that calls with the same fail_loc wakes up
* the first and continues. */
-#define OBD_RACE(id) \
-do { \
- if (unlikely(OBD_FAIL_CHECK_ONCE(id))) { \
- obd_race_state = 0; \
- CERROR("obd_race id %x sleeping\n", (id)); \
- OBD_SLEEP_ON(obd_race_waitq, obd_race_state != 0); \
- CERROR("obd_fail_race id %x awake\n", (id)); \
- } else if ((obd_fail_loc & OBD_FAIL_MASK_LOC) == \
- ((id) & OBD_FAIL_MASK_LOC)) { \
- CERROR("obd_fail_race id %x waking\n", (id)); \
- obd_race_state = 1; \
- wake_up(&obd_race_waitq); \
- } \
-} while(0)
+static inline void obd_race(__u32 id)
+{
+ if (OBD_FAIL_PRECHECK(id)) {
+ if (unlikely(__obd_fail_check_set(id, 0, OBD_FAIL_LOC_NOSET))) {
+ obd_race_state = 0;
+ CERROR("obd_race id %x sleeping\n", id);
+ OBD_SLEEP_ON(obd_race_waitq, obd_race_state != 0);
+ CERROR("obd_fail_race id %x awake\n", id);
+ } else {
+ CERROR("obd_fail_race id %x waking\n", id);
+ obd_race_state = 1;
+ wake_up(&obd_race_waitq);
+ }
+ }
+}
+#define OBD_RACE(id) obd_race(id)
#else
/* sigh. an expedient fix until OBD_RACE is fixed up */
#define OBD_RACE(foo) do {} while(0)
extern atomic_t libcfs_kmemory;
+#ifdef LPROCFS
+#define obd_memory_add(size) \
+ lprocfs_counter_add(obd_memory, OBD_MEMORY_STAT, (long)(size))
+#define obd_memory_sub(size) \
+ lprocfs_counter_sub(obd_memory, OBD_MEMORY_STAT, (long)(size))
+#define obd_memory_sum() \
+ lprocfs_stats_collector(obd_memory, OBD_MEMORY_STAT, \
+ LPROCFS_FIELDS_FLAGS_SUM)
+#define obd_pages_add(order) \
+ lprocfs_counter_add(obd_memory, OBD_MEMORY_PAGES_STAT, \
+ (long)(1 << (order)))
+#define obd_pages_sub(order) \
+ lprocfs_counter_sub(obd_memory, OBD_MEMORY_PAGES_STAT, \
+ (long)(1 << (order)))
+#define obd_pages_sum() \
+ lprocfs_stats_collector(obd_memory, OBD_MEMORY_PAGES_STAT, \
+ LPROCFS_FIELDS_FLAGS_SUM)
+
+extern void obd_update_maxusage(void);
+extern __u64 obd_memory_max(void);
+extern __u64 obd_pages_max(void);
+
+#else
+
+extern __u64 obd_alloc;
+extern __u64 obd_pages;
+
+extern __u64 obd_max_alloc;
+extern __u64 obd_max_pages;
+
+static inline void obd_memory_add(long size)
+{
+ obd_alloc += size;
+ if (obd_alloc > obd_max_alloc)
+ obd_max_alloc = obd_alloc;
+}
+
+static inline void obd_memory_sub(long size)
+{
+ obd_alloc -= size;
+}
+
+static inline void obd_pages_add(int order)
+{
+ obd_pages += 1<< order;
+ if (obd_pages > obd_max_pages)
+ obd_max_pages = obd_pages;
+}
+
+static inline void obd_pages_sub(int order)
+{
+ obd_pages -= 1<< order;
+}
+
+#define obd_memory_sum() (obd_alloc)
+#define obd_pages_sum() (obd_pages)
+
+#define obd_memory_max() (obd_max_alloc)
+#define obd_pages_max() (obd_max_pages)
+
+#endif
+
#if defined (CONFIG_DEBUG_MEMORY) && defined(__KERNEL__)
#define OBD_MT_WRONG_SIZE (1 << 0)
#if OBD_DEBUG_MEMUSAGE
#define OBD_ALLOC_POST(ptr, size, name) \
- atomic_add(size, &obd_memory); \
- if (atomic_read(&obd_memory) > obd_memmax) \
- obd_memmax = atomic_read(&obd_memory); \
+ obd_memory_add(size); \
get_mem_track((ptr), (size), __FILE__, __LINE__); \
- CDEBUG(D_MALLOC, name " '" #ptr "': %d at %p (tot %d)\n", \
- (int)(size), ptr, atomic_read(&obd_memory))
+ CDEBUG(D_MALLOC, name " '" #ptr "': %d at %p.\n", \
+ (int)(size), ptr)
#define OBD_FREE_PRE(ptr, size, name) \
LASSERT(ptr); \
put_mem_track((ptr), (size), __FILE__, __LINE__); \
- atomic_sub(size, &obd_memory); \
- CDEBUG(D_MALLOC, name " '" #ptr "': %d at %p (tot %d).\n", \
- (int)(size), ptr, atomic_read(&obd_memory)); \
+ obd_memory_sub(size); \
+ CDEBUG(D_MALLOC, name " '" #ptr "': %d at %p.\n", \
+ (int)(size), ptr); \
POISON(ptr, 0x5a, size)
#else /* !OBD_DEBUG_MEMUSAGE */
#endif /* !OBD_DEBUG_MEMUSAGE */
+#ifdef RANDOM_FAIL_ALLOC
+#define HAS_FAIL_ALLOC_FLAG OBD_FAIL_CHECK(OBD_FAIL_GENERAL_ALLOC)
+#else
+#define HAS_FAIL_ALLOC_FLAG 0
+#endif
+
+#define OBD_ALLOC_FAIL_BITS 24
+#define OBD_ALLOC_FAIL_MASK ((1 << OBD_ALLOC_FAIL_BITS) - 1)
+#define OBD_ALLOC_FAIL_MULT (OBD_ALLOC_FAIL_MASK / 100)
+
#if defined(LUSTRE_UTILS) /* this version is for utils only */
#define OBD_ALLOC_GFP(ptr, size, gfp_mask) \
do { \
} \
} while (0)
#else /* this version is for the kernel and liblustre */
+#define OBD_FREE_RTN0(ptr) \
+({ \
+ cfs_free(ptr); \
+ (ptr) = NULL; \
+ 0; \
+})
#define OBD_ALLOC_GFP(ptr, size, gfp_mask) \
do { \
(ptr) = cfs_alloc(size, (gfp_mask)); \
- if (unlikely((ptr) == NULL)) { \
- CERROR("kmalloc of '" #ptr "' (%d bytes) failed at %s:%d\n", \
- (int)(size), __FILE__, __LINE__); \
- CERROR("%d total bytes allocated by Lustre, %d by Portals\n", \
- atomic_read(&obd_memory), atomic_read(&libcfs_kmemory));\
- } else { \
+ if (likely((ptr) != NULL && \
+ (!HAS_FAIL_ALLOC_FLAG || obd_alloc_fail_rate == 0 || \
+ !obd_alloc_fail(ptr, #ptr, "km", size, \
+ __FILE__, __LINE__) || \
+ OBD_FREE_RTN0(ptr)))){ \
memset(ptr, 0, size); \
OBD_ALLOC_POST(ptr, size, "kmalloced"); \
} \
do { \
(ptr) = cfs_alloc_large(size); \
if (unlikely((ptr) == NULL)) { \
- CERROR("vmalloc of '" #ptr "' (%d bytes) failed at %s:%d\n", \
- (int)(size), __FILE__, __LINE__); \
- CERROR("%d total bytes allocated by Lustre, %d by Portals\n", \
- atomic_read(&obd_memory), atomic_read(&libcfs_kmemory));\
+ CERROR("vmalloc of '" #ptr "' (%d bytes) failed\n", \
+ (int)(size)); \
+ CERROR(LPU64" total bytes allocated by Lustre, %d by LNET\n", \
+ obd_memory_sum(), atomic_read(&libcfs_kmemory)); \
} else { \
memset(ptr, 0, size); \
OBD_ALLOC_POST(ptr, size, "vmalloced"); \
/* we memset() the slab object to 0 when allocation succeeds, so DO NOT
* HAVE A CTOR THAT DOES ANYTHING. its work will be cleared here. we'd
* love to assert on that, but slab.c keeps kmem_cache_s all to itself. */
+#define OBD_SLAB_FREE_RTN0(ptr, slab) \
+({ \
+ cfs_mem_cache_free((slab), (ptr)); \
+ (ptr) = NULL; \
+ 0; \
+})
#define OBD_SLAB_ALLOC(ptr, slab, type, size) \
do { \
LASSERT(!in_interrupt()); \
(ptr) = cfs_mem_cache_alloc(slab, (type)); \
- if (unlikely((ptr) == NULL)) { \
- CERROR("slab-alloc of '"#ptr"' (%d bytes) failed at %s:%d\n", \
- (int)(size), __FILE__, __LINE__); \
- CERROR("%d total bytes allocated by Lustre, %d by Portals\n", \
- atomic_read(&obd_memory), atomic_read(&libcfs_kmemory));\
- } else { \
+ if (likely((ptr) != NULL && \
+ (!HAS_FAIL_ALLOC_FLAG || obd_alloc_fail_rate == 0 || \
+ !obd_alloc_fail(ptr, #ptr, "slab-", size, \
+ __FILE__, __LINE__) || \
+ OBD_SLAB_FREE_RTN0(ptr, slab)))) { \
memset(ptr, 0, size); \
OBD_ALLOC_POST(ptr, size, "slab-alloced"); \
} \
#define OBD_SLAB_FREE_PTR(ptr, slab) \
OBD_SLAB_FREE((ptr), (slab), sizeof *(ptr))
-#define KEY_IS(str) (keylen >= strlen(str) && strcmp(key, str) == 0)
+#define KEY_IS(str) \
+ (keylen >= (sizeof(str)-1) && memcmp(key, str, (sizeof(str)-1)) == 0)
-#if defined(__linux__)
-#include <linux/obd_support.h>
-#elif defined(__APPLE__)
-#include <darwin/obd_support.h>
-#elif defined(__WINNT__)
-#include <winnt/obd_support.h>
-#else
-#error Unsupported operating system.
-#endif
+/* Wrapper for contiguous page frame allocation */
+#define OBD_PAGES_ALLOC(ptr, order, gfp_mask) \
+do { \
+ (ptr) = cfs_alloc_pages(gfp_mask, order); \
+ if (unlikely((ptr) == NULL)) { \
+ CERROR("alloc_pages of '" #ptr "' %d page(s) / "LPU64" bytes "\
+ "failed\n", (int)(1 << (order)), \
+ (__u64)((1 << (order)) << CFS_PAGE_SHIFT)); \
+ CERROR(LPU64" total bytes and "LPU64" total pages " \
+ "("LPU64" bytes) allocated by Lustre, " \
+ "%d total bytes by LNET\n", \
+ obd_memory_sum(), \
+ obd_pages_sum() << CFS_PAGE_SHIFT, \
+ obd_pages_sum(), \
+ atomic_read(&libcfs_kmemory)); \
+ } else { \
+ obd_pages_add(order); \
+ CDEBUG(D_MALLOC, "alloc_pages '" #ptr "': %d page(s) / " \
+ LPU64" bytes at %p.\n", \
+ (int)(1 << (order)), \
+ (__u64)((1 << (order)) << CFS_PAGE_SHIFT), ptr); \
+ } \
+} while (0)
+
+#define OBD_PAGE_ALLOC(ptr, gfp_mask) \
+ OBD_PAGES_ALLOC(ptr, 0, gfp_mask)
+
+#define OBD_PAGES_FREE(ptr, order) \
+do { \
+ LASSERT(ptr); \
+ obd_pages_sub(order); \
+ CDEBUG(D_MALLOC, "free_pages '" #ptr "': %d page(s) / "LPU64" bytes " \
+ "at %p.\n", \
+ (int)(1 << (order)), (__u64)((1 << (order)) << CFS_PAGE_SHIFT),\
+ ptr); \
+ __cfs_free_pages(ptr, order); \
+ (ptr) = (void *)0xdeadbeef; \
+} while (0)
+
+#define OBD_PAGE_FREE(ptr) OBD_PAGES_FREE(ptr, 0)
#endif