X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fptlrpc%2Fsec_bulk.c;h=1ddc335bb8a8ceed41da4c97ecb69673fd6f057b;hb=46bec835ac726aa2d06024b197002be7b112e192;hp=1e7f246a4a024b066d6ebe2f54015f9d4652976c;hpb=9c6ebfd11be1cc11629a2960c8c818492e1f0980;p=fs%2Flustre-release.git diff --git a/lustre/ptlrpc/sec_bulk.c b/lustre/ptlrpc/sec_bulk.c index 1e7f246..1ddc335 100644 --- a/lustre/ptlrpc/sec_bulk.c +++ b/lustre/ptlrpc/sec_bulk.c @@ -41,12 +41,6 @@ #define DEBUG_SUBSYSTEM S_SEC #include -#ifndef __KERNEL__ -#include -#include -#else -#include -#endif #include #include @@ -63,7 +57,6 @@ * bulk encryption page pools * ****************************************/ -#ifdef __KERNEL__ #define PTRS_PER_PAGE (PAGE_CACHE_SIZE / sizeof(void *)) #define PAGES_PER_POOL (PTRS_PER_PAGE) @@ -80,13 +73,13 @@ static struct ptlrpc_enc_page_pool { unsigned long epp_max_pages; /* maximum pages can hold, const */ unsigned int epp_max_pools; /* number of pools, const */ - /* - * wait queue in case of not enough free pages. - */ - cfs_waitq_t epp_waitq; /* waiting threads */ - unsigned int epp_waitqlen; /* wait queue length */ - unsigned long epp_pages_short; /* # of pages wanted of in-q users */ - unsigned int epp_growing:1; /* during adding pages */ + /* + * wait queue in case of not enough free pages. + */ + wait_queue_head_t epp_waitq; /* waiting threads */ + unsigned int epp_waitqlen; /* wait queue length */ + unsigned long epp_pages_short; /* # of pages wanted of in-q users */ + unsigned int epp_growing:1; /* during adding pages */ /* * indicating how idle the pools are, from 0 to MAX_IDLE_IDX @@ -136,14 +129,13 @@ static struct shrinker *pools_shrinker; /* * /proc/fs/lustre/sptlrpc/encrypt_page_pools */ -int sptlrpc_proc_read_enc_pool(char *page, char **start, off_t off, int count, - int *eof, void *data) +int sptlrpc_proc_enc_pool_seq_show(struct seq_file *m, void *v) { int rc; spin_lock(&page_pools.epp_lock); - rc = snprintf(page, count, + rc = seq_printf(m, "physical pages: %lu\n" "pages per pool: %lu\n" "max pages: %lu\n" @@ -163,7 +155,7 @@ int sptlrpc_proc_read_enc_pool(char *page, char **start, off_t off, int count, "max waitqueue depth: %u\n" "max wait time: "CFS_TIME_T"/%u\n" , - num_physpages, + totalram_pages, PAGES_PER_POOL, page_pools.epp_max_pages, page_pools.epp_max_pools, @@ -177,11 +169,11 @@ int sptlrpc_proc_read_enc_pool(char *page, char **start, off_t off, int count, page_pools.epp_st_grow_fails, page_pools.epp_st_shrinks, page_pools.epp_st_access, - page_pools.epp_st_missings, - page_pools.epp_st_lowfree, - page_pools.epp_st_max_wqlen, - page_pools.epp_st_max_wait, CFS_HZ - ); + page_pools.epp_st_missings, + page_pools.epp_st_lowfree, + page_pools.epp_st_max_wqlen, + page_pools.epp_st_max_wait, HZ + ); spin_unlock(&page_pools.epp_lock); return rc; @@ -221,7 +213,7 @@ static void enc_pools_release_free_pages(long npages) p_idx++; g_idx = 0; } - }; + } /* free unused pools */ while (p_idx_max1 < p_idx_max2) { @@ -233,30 +225,46 @@ static void enc_pools_release_free_pages(long npages) } /* - * could be called frequently for query (@nr_to_scan == 0). * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool. */ -static int enc_pools_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask)) +static unsigned long enc_pools_shrink_count(struct shrinker *s, + struct shrink_control *sc) { - if (unlikely(shrink_param(sc, nr_to_scan) != 0)) { + /* + * if no pool access for a long time, we consider it's fully idle. + * a little race here is fine. + */ + if (unlikely(cfs_time_current_sec() - page_pools.epp_last_access > + CACHE_QUIESCENT_PERIOD)) { spin_lock(&page_pools.epp_lock); - shrink_param(sc, nr_to_scan) = min_t(unsigned long, - shrink_param(sc, nr_to_scan), - page_pools.epp_free_pages - - PTLRPC_MAX_BRW_PAGES); - if (shrink_param(sc, nr_to_scan) > 0) { - enc_pools_release_free_pages(shrink_param(sc, - nr_to_scan)); - CDEBUG(D_SEC, "released %ld pages, %ld left\n", - (long)shrink_param(sc, nr_to_scan), - page_pools.epp_free_pages); - - page_pools.epp_st_shrinks++; - page_pools.epp_last_shrink = cfs_time_current_sec(); - } + page_pools.epp_idle_idx = IDLE_IDX_MAX; spin_unlock(&page_pools.epp_lock); } + LASSERT(page_pools.epp_idle_idx <= IDLE_IDX_MAX); + return max((int)page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES, 0) * + (IDLE_IDX_MAX - page_pools.epp_idle_idx) / IDLE_IDX_MAX; +} + +/* + * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool. + */ +static unsigned long enc_pools_shrink_scan(struct shrinker *s, + struct shrink_control *sc) +{ + spin_lock(&page_pools.epp_lock); + sc->nr_to_scan = min_t(unsigned long, sc->nr_to_scan, + page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES); + if (sc->nr_to_scan > 0) { + enc_pools_release_free_pages(sc->nr_to_scan); + CDEBUG(D_SEC, "released %ld pages, %ld left\n", + (long)sc->nr_to_scan, page_pools.epp_free_pages); + + page_pools.epp_st_shrinks++; + page_pools.epp_last_shrink = cfs_time_current_sec(); + } + spin_unlock(&page_pools.epp_lock); + /* * if no pool access for a long time, we consider it's fully idle. * a little race here is fine. @@ -269,10 +277,31 @@ static int enc_pools_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask)) } LASSERT(page_pools.epp_idle_idx <= IDLE_IDX_MAX); - return max((int)page_pools.epp_free_pages - PTLRPC_MAX_BRW_PAGES, 0) * - (IDLE_IDX_MAX - page_pools.epp_idle_idx) / IDLE_IDX_MAX; + return sc->nr_to_scan; } +#ifndef HAVE_SHRINKER_COUNT +/* + * could be called frequently for query (@nr_to_scan == 0). + * we try to keep at least PTLRPC_MAX_BRW_PAGES pages in the pool. + */ +static int enc_pools_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask)) +{ + struct shrink_control scv = { + .nr_to_scan = shrink_param(sc, nr_to_scan), + .gfp_mask = shrink_param(sc, gfp_mask) + }; +#if !defined(HAVE_SHRINKER_WANT_SHRINK_PTR) && !defined(HAVE_SHRINK_CONTROL) + struct shrinker* shrinker = NULL; +#endif + + enc_pools_shrink_scan(shrinker, &scv); + + return enc_pools_shrink_count(shrinker, &scv); +} + +#endif /* HAVE_SHRINKER_COUNT */ + static inline int npages_to_npools(unsigned long npages) { @@ -420,8 +449,8 @@ static int enc_pools_add_pages(int npages) goto out_pools; for (j = 0; j < PAGES_PER_POOL && alloced < npages; j++) { - pools[i][j] = alloc_page(__GFP_IO | - __GFP_HIGHMEM); + pools[i][j] = alloc_page(GFP_NOFS | + __GFP_HIGHMEM); if (pools[i][j] == NULL) goto out_pools; @@ -449,12 +478,11 @@ out: static inline void enc_pools_wakeup(void) { - LASSERT(spin_is_locked(&page_pools.epp_lock)); - LASSERT(page_pools.epp_waitqlen >= 0); + assert_spin_locked(&page_pools.epp_lock); if (unlikely(page_pools.epp_waitqlen)) { - LASSERT(cfs_waitq_active(&page_pools.epp_waitq)); - cfs_waitq_broadcast(&page_pools.epp_waitq); + LASSERT(waitqueue_active(&page_pools.epp_waitq)); + wake_up_all(&page_pools.epp_waitq); } } @@ -495,72 +523,72 @@ static int enc_pools_should_grow(int page_needed, long now) */ int sptlrpc_enc_pool_get_pages(struct ptlrpc_bulk_desc *desc) { - cfs_waitlink_t waitlink; - unsigned long this_idle = -1; - cfs_time_t tick = 0; - long now; - int p_idx, g_idx; - int i; + wait_queue_t waitlink; + unsigned long this_idle = -1; + cfs_time_t tick = 0; + long now; + int p_idx, g_idx; + int i; - LASSERT(desc->bd_iov_count > 0); - LASSERT(desc->bd_iov_count <= page_pools.epp_max_pages); + LASSERT(desc->bd_iov_count > 0); + LASSERT(desc->bd_iov_count <= page_pools.epp_max_pages); - /* resent bulk, enc iov might have been allocated previously */ - if (desc->bd_enc_iov != NULL) - return 0; + /* resent bulk, enc iov might have been allocated previously */ + if (desc->bd_enc_iov != NULL) + return 0; - OBD_ALLOC(desc->bd_enc_iov, - desc->bd_iov_count * sizeof(*desc->bd_enc_iov)); - if (desc->bd_enc_iov == NULL) - return -ENOMEM; + OBD_ALLOC(desc->bd_enc_iov, + desc->bd_iov_count * sizeof(*desc->bd_enc_iov)); + if (desc->bd_enc_iov == NULL) + return -ENOMEM; spin_lock(&page_pools.epp_lock); - page_pools.epp_st_access++; + page_pools.epp_st_access++; again: - if (unlikely(page_pools.epp_free_pages < desc->bd_iov_count)) { - if (tick == 0) - tick = cfs_time_current(); + if (unlikely(page_pools.epp_free_pages < desc->bd_iov_count)) { + if (tick == 0) + tick = cfs_time_current(); - now = cfs_time_current_sec(); + now = cfs_time_current_sec(); - page_pools.epp_st_missings++; - page_pools.epp_pages_short += desc->bd_iov_count; + page_pools.epp_st_missings++; + page_pools.epp_pages_short += desc->bd_iov_count; - if (enc_pools_should_grow(desc->bd_iov_count, now)) { - page_pools.epp_growing = 1; + if (enc_pools_should_grow(desc->bd_iov_count, now)) { + page_pools.epp_growing = 1; spin_unlock(&page_pools.epp_lock); enc_pools_add_pages(page_pools.epp_pages_short / 2); spin_lock(&page_pools.epp_lock); - page_pools.epp_growing = 0; + page_pools.epp_growing = 0; - enc_pools_wakeup(); - } else { - if (++page_pools.epp_waitqlen > - page_pools.epp_st_max_wqlen) - page_pools.epp_st_max_wqlen = - page_pools.epp_waitqlen; + enc_pools_wakeup(); + } else { + if (++page_pools.epp_waitqlen > + page_pools.epp_st_max_wqlen) + page_pools.epp_st_max_wqlen = + page_pools.epp_waitqlen; - cfs_set_current_state(CFS_TASK_UNINT); - cfs_waitlink_init(&waitlink); - cfs_waitq_add(&page_pools.epp_waitq, &waitlink); + set_current_state(TASK_UNINTERRUPTIBLE); + init_waitqueue_entry_current(&waitlink); + add_wait_queue(&page_pools.epp_waitq, &waitlink); spin_unlock(&page_pools.epp_lock); - cfs_waitq_wait(&waitlink, CFS_TASK_UNINT); - cfs_waitq_del(&page_pools.epp_waitq, &waitlink); + waitq_wait(&waitlink, TASK_UNINTERRUPTIBLE); + remove_wait_queue(&page_pools.epp_waitq, &waitlink); LASSERT(page_pools.epp_waitqlen > 0); spin_lock(&page_pools.epp_lock); - page_pools.epp_waitqlen--; - } + page_pools.epp_waitqlen--; + } - LASSERT(page_pools.epp_pages_short >= desc->bd_iov_count); - page_pools.epp_pages_short -= desc->bd_iov_count; + LASSERT(page_pools.epp_pages_short >= desc->bd_iov_count); + page_pools.epp_pages_short -= desc->bd_iov_count; - this_idle = 0; - goto again; - } + this_idle = 0; + goto again; + } /* record max wait time */ if (unlikely(tick != 0)) { @@ -708,16 +736,18 @@ static inline void enc_pools_free(void) int sptlrpc_enc_pool_init(void) { - /* - * maximum capacity is 1/8 of total physical memory. - * is the 1/8 a good number? - */ - page_pools.epp_max_pages = num_physpages / 8; - page_pools.epp_max_pools = npages_to_npools(page_pools.epp_max_pages); + DEF_SHRINKER_VAR(shvar, enc_pools_shrink, + enc_pools_shrink_count, enc_pools_shrink_scan); + /* + * maximum capacity is 1/8 of total physical memory. + * is the 1/8 a good number? + */ + page_pools.epp_max_pages = totalram_pages / 8; + page_pools.epp_max_pools = npages_to_npools(page_pools.epp_max_pages); - cfs_waitq_init(&page_pools.epp_waitq); - page_pools.epp_waitqlen = 0; - page_pools.epp_pages_short = 0; + init_waitqueue_head(&page_pools.epp_waitq); + page_pools.epp_waitqlen = 0; + page_pools.epp_pages_short = 0; page_pools.epp_growing = 0; @@ -743,8 +773,7 @@ int sptlrpc_enc_pool_init(void) if (page_pools.epp_pools == NULL) return -ENOMEM; - pools_shrinker = set_shrinker(pools_shrinker_seeks, - enc_pools_shrink); + pools_shrinker = set_shrinker(pools_shrinker_seeks, &shvar); if (pools_shrinker == NULL) { enc_pools_free(); return -ENOMEM; @@ -776,32 +805,12 @@ void sptlrpc_enc_pool_fini(void) CFS_TIME_T"/%d\n", page_pools.epp_st_max_pages, page_pools.epp_st_grows, page_pools.epp_st_grow_fails, - page_pools.epp_st_shrinks, page_pools.epp_st_access, - page_pools.epp_st_missings, page_pools.epp_st_max_wqlen, - page_pools.epp_st_max_wait, CFS_HZ); - } -} - -#else /* !__KERNEL__ */ - -int sptlrpc_enc_pool_get_pages(struct ptlrpc_bulk_desc *desc) -{ - return 0; -} - -void sptlrpc_enc_pool_put_pages(struct ptlrpc_bulk_desc *desc) -{ -} - -int sptlrpc_enc_pool_init(void) -{ - return 0; + page_pools.epp_st_shrinks, page_pools.epp_st_access, + page_pools.epp_st_missings, page_pools.epp_st_max_wqlen, + page_pools.epp_st_max_wait, HZ); + } } -void sptlrpc_enc_pool_fini(void) -{ -} -#endif static int cfs_hash_alg_id[] = { [BULK_HASH_ALG_NULL] = CFS_HASH_ALG_NULL, @@ -863,12 +872,17 @@ int bulk_sec_desc_unpack(struct lustre_msg *msg, int offset, int swabbed) } EXPORT_SYMBOL(bulk_sec_desc_unpack); +/* + * Compute the checksum of an RPC buffer payload. If the return \a buflen + * is not large enough, truncate the result to fit so that it is possible + * to use a hash function with a large hash space, but only use a part of + * the resulting hash. + */ int sptlrpc_get_bulk_checksum(struct ptlrpc_bulk_desc *desc, __u8 alg, void *buf, int buflen) { struct cfs_crypto_hash_desc *hdesc; int hashsize; - char hashbuf[64]; unsigned int bufsize; int i, err; @@ -885,28 +899,24 @@ int sptlrpc_get_bulk_checksum(struct ptlrpc_bulk_desc *desc, __u8 alg, hashsize = cfs_crypto_hash_digestsize(cfs_hash_alg_id[alg]); for (i = 0; i < desc->bd_iov_count; i++) { -#ifdef __KERNEL__ cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page, desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK, desc->bd_iov[i].kiov_len); -#else - cfs_crypto_hash_update(hdesc, desc->bd_iov[i].iov_base, - desc->bd_iov[i].iov_len); -#endif } + if (hashsize > buflen) { + unsigned char hashbuf[CFS_CRYPTO_HASH_DIGESTSIZE_MAX]; + bufsize = sizeof(hashbuf); - err = cfs_crypto_hash_final(hdesc, (unsigned char *)hashbuf, - &bufsize); + LASSERTF(bufsize >= hashsize, "bufsize = %u < hashsize %u\n", + bufsize, hashsize); + err = cfs_crypto_hash_final(hdesc, hashbuf, &bufsize); memcpy(buf, hashbuf, buflen); } else { bufsize = buflen; - err = cfs_crypto_hash_final(hdesc, (unsigned char *)buf, - &bufsize); + err = cfs_crypto_hash_final(hdesc, buf, &bufsize); } - if (err) - cfs_crypto_hash_final(hdesc, NULL, NULL); return err; } EXPORT_SYMBOL(sptlrpc_get_bulk_checksum);