/** * Copyright (c) 2016 - 2020 Tino Reichardt * All rights reserved. * * This source code is licensed under the BSD-style license found in the * LICENSE file in the root directory of this source tree. An additional grant * of patent rights can be found in the PATENTS file in the same directory. * * You can contact the author at: * - zstdmt source repository: https://github.com/mcmilk/zstdmt */ #include #include #define LZ4F_DISABLE_OBSOLETE_ENUMS #include "lz4frame.h" #include "memmt.h" #include "threading.h" #include "list.h" #include "lz4-mt.h" /** * multi threaded lz4 - multiple workers version * * - each thread works on his own * - no main thread which does reading and then starting the work * - needs a callback for reading / writing * - each worker does his: * 1) get read mutex and read some input * 2) release read mutex and do compression * 3) get write mutex and write result * 4) begin with step 1 again, until no input */ /* worker for compression */ typedef struct { LZ4MT_DCtx *ctx; pthread_t pthread; LZ4MT_Buffer in; LZ4F_decompressionContext_t dctx; } cwork_t; struct writelist; struct writelist { size_t frame; LZ4MT_Buffer out; struct list_head node; }; struct LZ4MT_DCtx_s { /* threads: 1..LZ4MT_THREAD_MAX */ int threads; /* should be used for read from input */ size_t inputsize; /* statistic */ size_t insize; size_t outsize; size_t curframe; size_t frames; /* threading */ cwork_t *cwork; /* reading input */ pthread_mutex_t read_mutex; fn_read *fn_read; void *arg_read; /* writing output */ pthread_mutex_t write_mutex; fn_write *fn_write; void *arg_write; /* lists for writing queue */ struct list_head writelist_free; struct list_head writelist_busy; struct list_head writelist_done; }; /* ************************************** * Decompression ****************************************/ LZ4MT_DCtx *LZ4MT_createDCtx(int threads, int inputsize) { LZ4MT_DCtx *ctx; int t; /* allocate ctx */ ctx = (LZ4MT_DCtx *) malloc(sizeof(LZ4MT_DCtx)); if (!ctx) return 0; /* check threads value */ if (threads < 1 || threads > LZ4MT_THREAD_MAX) return 0; /* setup ctx */ ctx->threads = threads; ctx->insize = 0; ctx->outsize = 0; ctx->frames = 0; ctx->curframe = 0; /* will be used for single stream only */ if (inputsize) ctx->inputsize = inputsize; else ctx->inputsize = 1024 + 1024 * 4; pthread_mutex_init(&ctx->read_mutex, NULL); pthread_mutex_init(&ctx->write_mutex, NULL); INIT_LIST_HEAD(&ctx->writelist_free); INIT_LIST_HEAD(&ctx->writelist_busy); INIT_LIST_HEAD(&ctx->writelist_done); ctx->cwork = (cwork_t *) malloc(sizeof(cwork_t) * threads); if (!ctx->cwork) goto err_cwork; for (t = 0; t < threads; t++) { cwork_t *w = &ctx->cwork[t]; w->ctx = ctx; /* setup thread work */ LZ4F_createDecompressionContext(&w->dctx, LZ4F_VERSION); } return ctx; err_cwork: free(ctx); return 0; } /** * mt_error - return mt lib specific error code */ static size_t mt_error(int rv) { switch (rv) { case -1: return ERROR(read_fail); case -2: return ERROR(canceled); case -3: return ERROR(memory_allocation); } /* XXX, some catch all other errors */ return ERROR(read_fail); } /** * pt_write - queue for decompressed output */ static size_t pt_write(LZ4MT_DCtx * ctx, struct writelist *wl) { struct list_head *entry; /* move the entry to the done list */ list_move(&wl->node, &ctx->writelist_done); again: /* check, what can be written ... */ list_for_each(entry, &ctx->writelist_done) { wl = list_entry(entry, struct writelist, node); if (wl->frame == ctx->curframe) { int rv = ctx->fn_write(ctx->arg_write, &wl->out); if (rv != 0) return mt_error(rv); ctx->outsize += wl->out.size; ctx->curframe++; list_move(entry, &ctx->writelist_free); goto again; } } return 0; } /** * pt_read - read compressed output */ static size_t pt_read(LZ4MT_DCtx * ctx, LZ4MT_Buffer * in, size_t * frame) { unsigned char hdrbuf[12]; LZ4MT_Buffer hdr; int rv; /* read skippable frame (8 or 12 bytes) */ pthread_mutex_lock(&ctx->read_mutex); /* special case, first 4 bytes already read */ if (ctx->frames == 0) { hdr.buf = hdrbuf + 4; hdr.size = 8; rv = ctx->fn_read(ctx->arg_read, &hdr); if (rv != 0) { pthread_mutex_unlock(&ctx->read_mutex); return mt_error(rv); } if (hdr.size != 8) goto error_read; hdr.buf = hdrbuf; } else { hdr.buf = hdrbuf; hdr.size = 12; rv = ctx->fn_read(ctx->arg_read, &hdr); if (rv != 0) { pthread_mutex_unlock(&ctx->read_mutex); return mt_error(rv); } /* eof reached ? */ if (hdr.size == 0) { pthread_mutex_unlock(&ctx->read_mutex); in->size = 0; return 0; } if (hdr.size != 12) goto error_read; if (MEM_readLE32((unsigned char *)hdr.buf + 0) != LZ4FMT_MAGIC_SKIPPABLE) goto error_data; } /* check header data */ if (MEM_readLE32((unsigned char *)hdr.buf + 4) != 4) goto error_data; ctx->insize += 12; /* read new inputsize */ { size_t toRead = MEM_readLE32((unsigned char *)hdr.buf + 8); if (in->allocated < toRead) { /* need bigger input buffer */ if (in->allocated) in->buf = realloc(in->buf, toRead); else in->buf = malloc(toRead); if (!in->buf) goto error_nomem; in->allocated = toRead; } in->size = toRead; rv = ctx->fn_read(ctx->arg_read, in); /* generic read failure! */ if (rv != 0) { pthread_mutex_unlock(&ctx->read_mutex); return mt_error(rv); } /* needed more bytes! */ if (in->size != toRead) goto error_data; ctx->insize += in->size; } *frame = ctx->frames++; pthread_mutex_unlock(&ctx->read_mutex); /* done, no error */ return 0; error_data: pthread_mutex_unlock(&ctx->read_mutex); return ERROR(data_error); error_read: pthread_mutex_unlock(&ctx->read_mutex); return ERROR(read_fail); error_nomem: pthread_mutex_unlock(&ctx->read_mutex); return ERROR(memory_allocation); } static void *pt_decompress(void *arg) { cwork_t *w = (cwork_t *) arg; LZ4MT_Buffer *in = &w->in; LZ4MT_DCtx *ctx = w->ctx; size_t result = 0; struct writelist *wl; for (;;) { struct list_head *entry; LZ4MT_Buffer *out; /* allocate space for new output */ pthread_mutex_lock(&ctx->write_mutex); if (!list_empty(&ctx->writelist_free)) { /* take unused entry */ entry = list_first(&ctx->writelist_free); wl = list_entry(entry, struct writelist, node); list_move(entry, &ctx->writelist_busy); } else { /* allocate new one */ wl = (struct writelist *) malloc(sizeof(struct writelist)); if (!wl) { result = ERROR(memory_allocation); goto error_unlock; } wl->out.buf = 0; wl->out.size = 0; wl->out.allocated = 0; list_add(&wl->node, &ctx->writelist_busy); } pthread_mutex_unlock(&ctx->write_mutex); out = &wl->out; /* zero should not happen here! */ result = pt_read(ctx, in, &wl->frame); if (LZ4MT_isError(result)) { list_move(&wl->node, &ctx->writelist_free); goto error_lock; } if (in->size == 0) break; /* mininmal frame */ if (in->size < 40 && ctx->frames == 1) { out->size = 1024 * 64; } else { /* get frame size for output buffer */ unsigned char *src = (unsigned char *)in->buf + 6; out->size = (size_t) MEM_readLE64(src); } if (out->allocated < out->size) { if (out->allocated) out->buf = realloc(out->buf, out->size); else out->buf = malloc(out->size); if (!out->buf) { result = ERROR(memory_allocation); goto error_lock; } out->allocated = out->size; } result = LZ4F_decompress(w->dctx, out->buf, &out->size, in->buf, &in->size, 0); if (LZ4F_isError(result)) { lz4mt_errcode = result; result = ERROR(compression_library); goto error_lock; } if (result != 0) { result = ERROR(frame_decompress); goto error_lock; } /* write result */ pthread_mutex_lock(&ctx->write_mutex); result = pt_write(ctx, wl); if (LZ4MT_isError(result)) goto error_unlock; pthread_mutex_unlock(&ctx->write_mutex); } /* everything is okay */ pthread_mutex_lock(&ctx->write_mutex); list_move(&wl->node, &ctx->writelist_free); pthread_mutex_unlock(&ctx->write_mutex); if (in->allocated) free(in->buf); return 0; error_lock: pthread_mutex_lock(&ctx->write_mutex); error_unlock: list_move(&wl->node, &ctx->writelist_free); pthread_mutex_unlock(&ctx->write_mutex); if (in->allocated) free(in->buf); return (void *)result; } /* single threaded */ static size_t st_decompress(void *arg) { LZ4MT_DCtx *ctx = (LZ4MT_DCtx *) arg; LZ4F_errorCode_t result = 0; cwork_t *w = &ctx->cwork[0]; LZ4MT_Buffer Out; LZ4MT_Buffer *out = &Out; LZ4MT_Buffer *in = &w->in; void *magic = in->buf; int rv; /* allocate space for input buffer */ in->size = ctx->inputsize; in->buf = malloc(in->size); if (!in->buf) return ERROR(memory_allocation); /* allocate space for output buffer */ out->size = ctx->inputsize; out->buf = malloc(out->size); if (!out->buf) { free(in->buf); return ERROR(memory_allocation); } /* we have read already 4 bytes */ in->size = 4; memcpy(in->buf, magic, in->size); /* stats */ ctx->insize = 4; ctx->outsize = 0; /* decompress loop */ for (;;) { size_t srcPos = 0; for (;;) { size_t srcSize = in->size - srcPos; out->size = ctx->inputsize; result = LZ4F_decompress(w->dctx, out->buf, &out->size, (unsigned char *)in->buf + srcPos, &srcSize, NULL); if (LZ4F_isError(result)) { free(in->buf); free(out->buf); return ERROR(compression_library); } /* update stats */ srcPos += srcSize; ctx->insize += srcSize; ctx->outsize += out->size; /* have some output */ if (out->size) { rv = ctx->fn_write(ctx->arg_write, out); if (rv != 0) { free(in->buf); free(out->buf); return mt_error(rv); } } /* consumed all input */ if (srcPos == in->size) break; } /* read new input */ if (result) in->size = result; else in->size = ctx->inputsize; if (in->size > ctx->inputsize) in->size = ctx->inputsize; rv = ctx->fn_read(ctx->arg_read, in); ctx->insize += in->size; if (rv != 0) { free(in->buf); free(out->buf); return mt_error(rv); } if (in->size == 0) break; } /* no error */ free(out->buf); free(in->buf); return 0; } size_t LZ4MT_decompressDCtx(LZ4MT_DCtx * ctx, LZ4MT_RdWr_t * rdwr) { unsigned char buf[4]; int t, rv; cwork_t *w = &ctx->cwork[0]; LZ4MT_Buffer *in = &w->in; void *retval_of_thread = 0; if (!ctx) return ERROR(compressionParameter_unsupported); /* init reading and writing functions */ ctx->fn_read = rdwr->fn_read; ctx->fn_write = rdwr->fn_write; ctx->arg_read = rdwr->arg_read; ctx->arg_write = rdwr->arg_write; /* check for LZ4FMT_MAGIC_SKIPPABLE */ in->buf = buf; in->size = 4; rv = ctx->fn_read(ctx->arg_read, in); if (rv != 0) return mt_error(rv); if (in->size != 4) return ERROR(data_error); /* single threaded with unknown sizes */ if (MEM_readLE32(buf) != LZ4FMT_MAGIC_SKIPPABLE) { /* look for correct magic */ if (MEM_readLE32(buf) != LZ4FMT_MAGICNUMBER) return ERROR(data_error); /* decompress single threaded */ return st_decompress(ctx); } /* mark unused */ in->buf = 0; in->size = 0; in->allocated = 0; /* single threaded, but with known sizes */ if (ctx->threads == 1) { /* no pthread_create() needed! */ void *p = pt_decompress(w); if (p) return (size_t) p; goto okay; } /* multi threaded */ for (t = 0; t < ctx->threads; t++) { cwork_t *wt = &ctx->cwork[t]; wt->in.buf = 0; wt->in.size = 0; wt->in.allocated = 0; pthread_create(&wt->pthread, NULL, pt_decompress, wt); } /* wait for all workers */ for (t = 0; t < ctx->threads; t++) { cwork_t *wt = &ctx->cwork[t]; void *p = 0; pthread_join(wt->pthread, &p); if (p) retval_of_thread = p; } okay: /* clean up the buffers */ while (!list_empty(&ctx->writelist_free)) { struct writelist *wl; struct list_head *entry; entry = list_first(&ctx->writelist_free); wl = list_entry(entry, struct writelist, node); free(wl->out.buf); list_del(&wl->node); free(wl); } return (size_t) retval_of_thread; } /* returns current uncompressed data size */ size_t LZ4MT_GetInsizeDCtx(LZ4MT_DCtx * ctx) { if (!ctx) return 0; return ctx->insize; } /* returns the current compressed data size */ size_t LZ4MT_GetOutsizeDCtx(LZ4MT_DCtx * ctx) { if (!ctx) return 0; return ctx->outsize; } /* returns the current compressed frames */ size_t LZ4MT_GetFramesDCtx(LZ4MT_DCtx * ctx) { if (!ctx) return 0; return ctx->curframe; } void LZ4MT_freeDCtx(LZ4MT_DCtx * ctx) { int t; if (!ctx) return; for (t = 0; t < ctx->threads; t++) { cwork_t *w = &ctx->cwork[t]; LZ4F_freeDecompressionContext(w->dctx); } pthread_mutex_destroy(&ctx->read_mutex); pthread_mutex_destroy(&ctx->write_mutex); free(ctx->cwork); free(ctx); ctx = 0; return; }