rand: add 64-bit tausworthe variant with a 2^258 cycle
[fio.git] / verify.c
index e59a4b29051065cd9ebd1caf0ab1b9494f313211..fcdf748810ea2634d0402588c546ff6d0a08b949 100644 (file)
--- a/verify.c
+++ b/verify.c
@@ -29,62 +29,47 @@ static void populate_hdr(struct thread_data *td, struct io_u *io_u,
                         struct verify_header *hdr, unsigned int header_num,
                         unsigned int header_len);
 
-static void fill_pattern(struct thread_data *td, void *p, unsigned int len,
-                        char *pattern, unsigned int pattern_bytes)
+void fill_buffer_pattern(struct thread_data *td, void *p, unsigned int len)
 {
-       switch (pattern_bytes) {
-       case 0:
-               assert(0);
-               break;
-       case 1:
-               dprint(FD_VERIFY, "fill verify pattern b=0 len=%u\n", len);
-               memset(p, pattern[0], len);
-               break;
-       default: {
-               unsigned int i = 0, size = 0;
-               unsigned char *b = p;
-
-               dprint(FD_VERIFY, "fill verify pattern b=%d len=%u\n",
-                                       pattern_bytes, len);
-
-               while (i < len) {
-                       size = pattern_bytes;
-                       if (size > (len - i))
-                               size = len - i;
-                       memcpy(b+i, pattern, size);
-                       i += size;
-               }
-               break;
-               }
-       }
+       fill_pattern(p, len, td->o.buffer_pattern, td->o.buffer_pattern_bytes);
 }
 
-void fill_buffer_pattern(struct thread_data *td, void *p, unsigned int len)
+void __fill_buffer(struct thread_options *o, unsigned long seed, void *p,
+                  unsigned int len)
 {
-       fill_pattern(td, p, len, td->o.buffer_pattern, td->o.buffer_pattern_bytes);
+       __fill_random_buf_percentage(seed, p, o->compress_percentage, len, len, o->buffer_pattern, o->buffer_pattern_bytes);
+}
+
+unsigned long fill_buffer(struct thread_data *td, void *p, unsigned int len)
+{
+       struct frand_state *fs = &td->verify_state;
+       struct thread_options *o = &td->o;
+
+       return fill_random_buf_percentage(fs, p, o->compress_percentage, len, len, o->buffer_pattern, o->buffer_pattern_bytes);
 }
 
 void fill_verify_pattern(struct thread_data *td, void *p, unsigned int len,
                         struct io_u *io_u, unsigned long seed, int use_seed)
 {
-       if (!td->o.verify_pattern_bytes) {
+       struct thread_options *o = &td->o;
+
+       if (!o->verify_pattern_bytes) {
                dprint(FD_VERIFY, "fill random bytes len=%u\n", len);
 
                if (use_seed)
-                       __fill_random_buf(p, len, seed);
+                       __fill_buffer(o, seed, p, len);
                else
-                       io_u->rand_seed = fill_random_buf(&td->__verify_state, p, len);
+                       io_u->rand_seed = fill_buffer(td, p, len);
                return;
        }
 
        if (io_u->buf_filled_len >= len) {
                dprint(FD_VERIFY, "using already filled verify pattern b=%d len=%u\n",
-                       td->o.verify_pattern_bytes, len);
+                       o->verify_pattern_bytes, len);
                return;
        }
 
-       fill_pattern(td, p, len, td->o.verify_pattern, td->o.verify_pattern_bytes);
-
+       fill_pattern(p, len, o->verify_pattern, o->verify_pattern_bytes);
        io_u->buf_filled_len = len;
 }
 
@@ -405,13 +390,14 @@ static int verify_io_u_meta(struct verify_header *hdr, struct vcont *vc)
 
        /*
         * For read-only workloads, the program cannot be certain of the
-        * last numberio written to a block. Checking of numberio will be done
-        * only for workloads that write data.
-        * For verify_only, numberio will be checked in the last iteration when
-        * the correct state of numberio, that would have been written to each
-        * block in a previous run of fio, has been reached.
+        * last numberio written to a block. Checking of numberio will be
+        * done only for workloads that write data.  For verify_only,
+        * numberio will be checked in the last iteration when the correct
+        * state of numberio, that would have been written to each block
+        * in a previous run of fio, has been reached.
         */
-       if (td_write(td) || td_rw(td))
+       if ((td_write(td) || td_rw(td)) && (td_min_bs(td) == td_max_bs(td)) &&
+           !td->o.time_based)
                if (!td->o.verify_only || td->o.loops == 0)
                        if (vh->numberio != io_u->numberio)
                                ret = EILSEQ;
@@ -486,6 +472,7 @@ static int verify_io_u_sha256(struct verify_header *hdr, struct vcont *vc)
 
        fio_sha256_init(&sha256_ctx);
        fio_sha256_update(&sha256_ctx, p, hdr->len - hdr_size(hdr));
+       fio_sha256_final(&sha256_ctx);
 
        if (!memcmp(vh->sha256, sha256_ctx.buf, sizeof(sha256)))
                return 0;
@@ -511,6 +498,7 @@ static int verify_io_u_sha1(struct verify_header *hdr, struct vcont *vc)
 
        fio_sha1_init(&sha1_ctx);
        fio_sha1_update(&sha1_ctx, p, hdr->len - hdr_size(hdr));
+       fio_sha1_final(&sha1_ctx);
 
        if (!memcmp(vh->sha1, sha1_ctx.H, sizeof(sha1)))
                return 0;
@@ -641,6 +629,7 @@ static int verify_io_u_md5(struct verify_header *hdr, struct vcont *vc)
 
        fio_md5_init(&md5_ctx);
        fio_md5_update(&md5_ctx, p, hdr->len - hdr_size(hdr));
+       fio_md5_final(&md5_ctx);
 
        if (!memcmp(vh->md5_digest, md5_ctx.hash, sizeof(hash)))
                return 0;
@@ -656,19 +645,21 @@ static int verify_io_u_md5(struct verify_header *hdr, struct vcont *vc)
 /*
  * Push IO verification to a separate thread
  */
-int verify_io_u_async(struct thread_data *td, struct io_u *io_u)
+int verify_io_u_async(struct thread_data *td, struct io_u **io_u_ptr)
 {
-       if (io_u->file)
-               put_file_log(td, io_u->file);
+       struct io_u *io_u = *io_u_ptr;
 
        pthread_mutex_lock(&td->io_u_lock);
 
+       if (io_u->file)
+               put_file_log(td, io_u->file);
+
        if (io_u->flags & IO_U_F_IN_CUR_DEPTH) {
                td->cur_depth--;
-               io_u->flags &= ~IO_U_F_IN_CUR_DEPTH;
+               io_u_clear(io_u, IO_U_F_IN_CUR_DEPTH);
        }
        flist_add_tail(&io_u->verify_list, &td->verify_list);
-       io_u->flags |= IO_U_F_FREE_DEF;
+       *io_u_ptr = NULL;
        pthread_mutex_unlock(&td->io_u_lock);
 
        pthread_cond_signal(&td->verify_cond);
@@ -747,15 +738,23 @@ err:
        return EILSEQ;
 }
 
-int verify_io_u(struct thread_data *td, struct io_u *io_u)
+int verify_io_u(struct thread_data *td, struct io_u **io_u_ptr)
 {
        struct verify_header *hdr;
+       struct io_u *io_u = *io_u_ptr;
        unsigned int header_size, hdr_inc, hdr_num = 0;
        void *p;
        int ret;
 
        if (td->o.verify == VERIFY_NULL || io_u->ddir != DDIR_READ)
                return 0;
+       /*
+        * If the IO engine is faking IO (like null), then just pretend
+        * we verified everything.
+        */
+       if (td->io_ops->flags & FIO_FAKEIO)
+               return 0;
+
        if (io_u->flags & IO_U_F_TRIMMED) {
                ret = verify_trimmed_io_u(td, io_u);
                goto done;
@@ -897,6 +896,7 @@ static void fill_sha256(struct verify_header *hdr, void *p, unsigned int len)
 
        fio_sha256_init(&sha256_ctx);
        fio_sha256_update(&sha256_ctx, p, len);
+       fio_sha256_final(&sha256_ctx);
 }
 
 static void fill_sha1(struct verify_header *hdr, void *p, unsigned int len)
@@ -908,6 +908,7 @@ static void fill_sha1(struct verify_header *hdr, void *p, unsigned int len)
 
        fio_sha1_init(&sha1_ctx);
        fio_sha1_update(&sha1_ctx, p, len);
+       fio_sha1_final(&sha1_ctx);
 }
 
 static void fill_crc7(struct verify_header *hdr, void *p, unsigned int len)
@@ -954,6 +955,7 @@ static void fill_md5(struct verify_header *hdr, void *p, unsigned int len)
 
        fio_md5_init(&md5_ctx);
        fio_md5_update(&md5_ctx, p, len);
+       fio_md5_final(&md5_ctx);
 }
 
 static void populate_hdr(struct thread_data *td, struct io_u *io_u,
@@ -1103,10 +1105,10 @@ int get_next_verify(struct thread_data *td, struct io_u *io_u)
                io_u->buflen = ipo->len;
                io_u->numberio = ipo->numberio;
                io_u->file = ipo->file;
-               io_u->flags |= IO_U_F_VER_LIST;
+               io_u_set(io_u, IO_U_F_VER_LIST);
 
                if (ipo->flags & IP_F_TRIMMED)
-                       io_u->flags |= IO_U_F_TRIMMED;
+                       io_u_set(io_u, IO_U_F_TRIMMED);
 
                if (!fio_file_open(io_u->file)) {
                        int r = td_io_open_file(td, io_u->file);
@@ -1129,9 +1131,9 @@ int get_next_verify(struct thread_data *td, struct io_u *io_u)
                dprint(FD_VERIFY, "get_next_verify: ret io_u %p\n", io_u);
 
                if (!td->o.verify_pattern_bytes) {
-                       io_u->rand_seed = __rand(&td->__verify_state);
+                       io_u->rand_seed = __rand(&td->verify_state);
                        if (sizeof(int) != sizeof(long *))
-                               io_u->rand_seed *= __rand(&td->__verify_state);
+                               io_u->rand_seed *= __rand(&td->verify_state);
                }
                return 0;
        }
@@ -1155,7 +1157,7 @@ static void *verify_async_thread(void *data)
        struct io_u *io_u;
        int ret = 0;
 
-       if (td->o.verify_cpumask_set &&
+       if (fio_option_is_set(&td->o, verify_cpumask) &&
            fio_setaffinity(td->pid, td->o.verify_cpumask)) {
                log_err("fio: failed setting verify thread affinity\n");
                goto done;
@@ -1188,9 +1190,11 @@ static void *verify_async_thread(void *data)
 
                while (!flist_empty(&list)) {
                        io_u = flist_first_entry(&list, struct io_u, verify_list);
-                       flist_del(&io_u->verify_list);
+                       flist_del_init(&io_u->verify_list);
+
+                       io_u_set(io_u, IO_U_F_NO_FILE_PUT);
+                       ret = verify_io_u(td, &io_u);
 
-                       ret = verify_io_u(td, io_u);
                        put_io_u(td, io_u);
                        if (!ret)
                                continue;
@@ -1273,3 +1277,357 @@ void verify_async_exit(struct thread_data *td)
        free(td->verify_threads);
        td->verify_threads = NULL;
 }
+
+struct all_io_list *get_all_io_list(int save_mask, size_t *sz)
+{
+       struct all_io_list *rep;
+       struct thread_data *td;
+       size_t depth;
+       void *next;
+       int i, nr;
+
+       compiletime_assert(sizeof(struct all_io_list) == 8, "all_io_list");
+
+       /*
+        * Calculate reply space needed. We need one 'io_state' per thread,
+        * and the size will vary depending on depth.
+        */
+       depth = 0;
+       nr = 0;
+       for_each_td(td, i) {
+               if (save_mask != IO_LIST_ALL && (i + 1) != save_mask)
+                       continue;
+               td->stop_io = 1;
+               td->flags |= TD_F_VSTATE_SAVED;
+               depth += td->o.iodepth;
+               nr++;
+       }
+
+       if (!nr)
+               return NULL;
+
+       *sz = sizeof(*rep);
+       *sz += nr * sizeof(struct thread_io_list);
+       *sz += depth * sizeof(uint64_t);
+       rep = malloc(*sz);
+
+       rep->threads = cpu_to_le64((uint64_t) nr);
+
+       next = &rep->state[0];
+       for_each_td(td, i) {
+               struct thread_io_list *s = next;
+               unsigned int comps;
+
+               if (save_mask != IO_LIST_ALL && (i + 1) != save_mask)
+                       continue;
+
+               if (td->last_write_comp) {
+                       int j, k;
+
+                       if (td->io_blocks[DDIR_WRITE] < td->o.iodepth)
+                               comps = td->io_blocks[DDIR_WRITE];
+                       else
+                               comps = td->o.iodepth;
+
+                       k = td->last_write_idx - 1;
+                       for (j = 0; j < comps; j++) {
+                               if (k == -1)
+                                       k = td->o.iodepth - 1;
+                               s->offsets[j] = cpu_to_le64(td->last_write_comp[k]);
+                               k--;
+                       }
+               } else
+                       comps = 0;
+
+               s->no_comps = cpu_to_le64((uint64_t) comps);
+               s->depth = cpu_to_le64((uint64_t) td->o.iodepth);
+               s->numberio = cpu_to_le64((uint64_t) td->io_issues[DDIR_WRITE]);
+               s->index = cpu_to_le64((uint64_t) i);
+               if (td->random_state.use64) {
+                       s->rand.state64.s[0] = cpu_to_le64(td->random_state.state64.s1);
+                       s->rand.state64.s[1] = cpu_to_le64(td->random_state.state64.s2);
+                       s->rand.state64.s[2] = cpu_to_le64(td->random_state.state64.s3);
+                       s->rand.state64.s[3] = cpu_to_le64(td->random_state.state64.s4);
+                       s->rand.state64.s[4] = cpu_to_le64(td->random_state.state64.s5);
+                       s->rand.state64.s[5] = 0;
+                       s->rand.use64 = cpu_to_le64((uint64_t)1);
+               } else {
+                       s->rand.state32.s[0] = cpu_to_le32(td->random_state.state32.s1);
+                       s->rand.state32.s[1] = cpu_to_le32(td->random_state.state32.s2);
+                       s->rand.state32.s[2] = cpu_to_le32(td->random_state.state32.s3);
+                       s->rand.state32.s[3] = 0;
+                       s->rand.use64 = 0;
+               }
+               s->name[sizeof(s->name) - 1] = '\0';
+               strncpy((char *) s->name, td->o.name, sizeof(s->name) - 1);
+               next = io_list_next(s);
+       }
+
+       return rep;
+}
+
+static int open_state_file(const char *name, const char *prefix, int num,
+                          int for_write)
+{
+       char out[64];
+       int flags;
+       int fd;
+
+       if (for_write)
+               flags = O_CREAT | O_TRUNC | O_WRONLY | O_SYNC;
+       else
+               flags = O_RDONLY;
+
+       verify_state_gen_name(out, sizeof(out), name, prefix, num);
+
+       fd = open(out, flags, 0644);
+       if (fd == -1) {
+               perror("fio: open state file");
+               return -1;
+       }
+
+       return fd;
+}
+
+static int write_thread_list_state(struct thread_io_list *s,
+                                  const char *prefix)
+{
+       struct verify_state_hdr hdr;
+       uint64_t crc;
+       ssize_t ret;
+       int fd;
+
+       fd = open_state_file((const char *) s->name, prefix, s->index, 1);
+       if (fd == -1)
+               return 1;
+
+       crc = fio_crc32c((void *)s, thread_io_list_sz(s));
+
+       hdr.version = cpu_to_le64((uint64_t) VSTATE_HDR_VERSION);
+       hdr.size = cpu_to_le64((uint64_t) thread_io_list_sz(s));
+       hdr.crc = cpu_to_le64(crc);
+       ret = write(fd, &hdr, sizeof(hdr));
+       if (ret != sizeof(hdr))
+               goto write_fail;
+
+       ret = write(fd, s, thread_io_list_sz(s));
+       if (ret != thread_io_list_sz(s)) {
+write_fail:
+               if (ret < 0)
+                       perror("fio: write state file");
+               log_err("fio: failed to write state file\n");
+               ret = 1;
+       } else
+               ret = 0;
+
+       close(fd);
+       return ret;
+}
+
+void __verify_save_state(struct all_io_list *state, const char *prefix)
+{
+       struct thread_io_list *s = &state->state[0];
+       unsigned int i;
+
+       for (i = 0; i < le64_to_cpu(state->threads); i++) {
+               write_thread_list_state(s,  prefix);
+               s = io_list_next(s);
+       }
+}
+
+void verify_save_state(void)
+{
+       struct all_io_list *state;
+       size_t sz;
+
+       state = get_all_io_list(IO_LIST_ALL, &sz);
+       if (state) {
+               __verify_save_state(state, "local");
+               free(state);
+       }
+}
+
+void verify_free_state(struct thread_data *td)
+{
+       if (td->vstate)
+               free(td->vstate);
+}
+
+static struct thread_io_list *convert_v1_list(struct thread_io_list_v1 *s)
+{
+       struct thread_io_list *til;
+       int i;
+
+       til = malloc(__thread_io_list_sz(s->no_comps));
+       til->no_comps = s->no_comps;
+       til->depth = s->depth;
+       til->numberio = s->numberio;
+       til->index = s->index;
+       memcpy(til->name, s->name, sizeof(til->name));
+
+       til->rand.use64 = 0;
+       for (i = 0; i < 4; i++)
+               til->rand.state32.s[i] = s->rand.s[i];
+
+       for (i = 0; i < s->no_comps; i++)
+               til->offsets[i] = s->offsets[i];
+
+       return til;
+}
+
+void verify_convert_assign_state(struct thread_data *td, void *p, int version)
+{
+       struct thread_io_list *til;
+       int i;
+
+       if (version == 1) {
+               struct thread_io_list_v1 *s = p;
+
+               s->no_comps = le64_to_cpu(s->no_comps);
+               s->depth = le64_to_cpu(s->depth);
+               s->numberio = le64_to_cpu(s->numberio);
+               for (i = 0; i < 4; i++)
+                       s->rand.s[i] = le32_to_cpu(s->rand.s[i]);
+               for (i = 0; i < s->no_comps; i++)
+                       s->offsets[i] = le64_to_cpu(s->offsets[i]);
+
+               til = convert_v1_list(s);
+               free(s);
+       } else {
+               struct thread_io_list *s = p;
+
+               s->no_comps = le64_to_cpu(s->no_comps);
+               s->depth = le64_to_cpu(s->depth);
+               s->numberio = le64_to_cpu(s->numberio);
+               s->rand.use64 = le64_to_cpu(s->rand.use64);
+
+               if (s->rand.use64) {
+                       for (i = 0; i < 6; i++)
+                               s->rand.state64.s[i] = le64_to_cpu(s->rand.state64.s[i]);
+               } else {
+                       for (i = 0; i < 4; i++)
+                               s->rand.state32.s[i] = le32_to_cpu(s->rand.state32.s[i]);
+               }
+               for (i = 0; i < s->no_comps; i++)
+                       s->offsets[i] = le64_to_cpu(s->offsets[i]);
+
+               til = p;
+       }
+
+       td->vstate = til;
+}
+
+int verify_state_hdr(struct verify_state_hdr *hdr, struct thread_io_list *s,
+                    int *version)
+{
+       uint64_t crc;
+
+       hdr->version = le64_to_cpu(hdr->version);
+       hdr->size = le64_to_cpu(hdr->size);
+       hdr->crc = le64_to_cpu(hdr->crc);
+
+       if (hdr->version != VSTATE_HDR_VERSION ||
+           hdr->version != VSTATE_HDR_VERSION_V1)
+               return 1;
+
+       crc = fio_crc32c((void *)s, hdr->size);
+       if (crc != hdr->crc)
+               return 1;
+
+       *version = hdr->version;
+       return 0;
+}
+
+int verify_load_state(struct thread_data *td, const char *prefix)
+{
+       struct verify_state_hdr hdr;
+       void *s = NULL;
+       uint64_t crc;
+       ssize_t ret;
+       int fd;
+
+       if (!td->o.verify_state)
+               return 0;
+
+       fd = open_state_file(td->o.name, prefix, td->thread_number - 1, 0);
+       if (fd == -1)
+               return 1;
+
+       ret = read(fd, &hdr, sizeof(hdr));
+       if (ret != sizeof(hdr)) {
+               if (ret < 0)
+                       td_verror(td, errno, "read verify state hdr");
+               log_err("fio: failed reading verify state header\n");
+               goto err;
+       }
+
+       hdr.version = le64_to_cpu(hdr.version);
+       hdr.size = le64_to_cpu(hdr.size);
+       hdr.crc = le64_to_cpu(hdr.crc);
+
+       if (hdr.version != VSTATE_HDR_VERSION &&
+           hdr.version != VSTATE_HDR_VERSION_V1) {
+               log_err("fio: bad version in verify state header\n");
+               goto err;
+       }
+
+       s = malloc(hdr.size);
+       ret = read(fd, s, hdr.size);
+       if (ret != hdr.size) {
+               if (ret < 0)
+                       td_verror(td, errno, "read verify state");
+               log_err("fio: failed reading verity state\n");
+               goto err;
+       }
+
+       crc = fio_crc32c(s, hdr.size);
+       if (crc != hdr.crc) {
+               log_err("fio: verify state is corrupt\n");
+               goto err;
+       }
+
+       close(fd);
+
+       verify_convert_assign_state(td, s, hdr.version);
+       return 0;
+err:
+       if (s)
+               free(s);
+       close(fd);
+       return 1;
+}
+
+/*
+ * Use the loaded verify state to know when to stop doing verification
+ */
+int verify_state_should_stop(struct thread_data *td, struct io_u *io_u)
+{
+       struct thread_io_list *s = td->vstate;
+       int i;
+
+       if (!s)
+               return 0;
+
+       /*
+        * If we're not into the window of issues - depth yet, continue. If
+        * issue is shorter than depth, do check.
+        */
+       if ((td->io_blocks[DDIR_READ] < s->depth ||
+           s->numberio - td->io_blocks[DDIR_READ] > s->depth) &&
+           s->numberio > s->depth)
+               return 0;
+
+       /*
+        * We're in the window of having to check if this io was
+        * completed or not. If the IO was seen as completed, then
+        * lets verify it.
+        */
+       for (i = 0; i < s->no_comps; i++)
+               if (io_u->offset == s->offsets[i])
+                       return 0;
+
+       /*
+        * Not found, we have to stop
+        */
+       return 1;
+}