X-Git-Url: https://git.octo.it/?p=rrdtool.git;a=blobdiff_plain;f=src%2Frrd_open.c;h=5f3723f30cef086d44c6bab4a3c434b1d74e6507;hp=c986f750beca7bec9fb3f0cba5c5a488cfda17d2;hb=9d7aa8ddae065b774fe4eb1df76a3242b9db2d53;hpb=009ac7ac9dde1ba7c450b230b9118cff2cedc9f6 diff --git a/src/rrd_open.c b/src/rrd_open.c index c986f75..5f3723f 100644 --- a/src/rrd_open.c +++ b/src/rrd_open.c @@ -1,149 +1,146 @@ /***************************************************************************** - * RRDtool 1.2.23 Copyright by Tobi Oetiker, 1997-2007 + * RRDtool 1.4.3 Copyright by Tobi Oetiker, 1997-2010 ***************************************************************************** * rrd_open.c Open an RRD File ***************************************************************************** * $Id$ - * $Log$ - * Revision 1.10 2004/05/26 22:11:12 oetiker - * reduce compiler warnings. Many small fixes. -- Mike Slifcak - * - * Revision 1.9 2003/04/29 21:56:49 oetiker - * readline in rrd_open.c reads the file in 8 KB blocks, and calls realloc for - * each block. realloc is very slow in Mac OS X for huge blocks, e.g. when - * restoring databases from huge xml files. This patch finds the size of the - * file, and starts out with malloc'ing the full size. - * -- Peter Speck - * - * Revision 1.8 2003/04/11 19:43:44 oetiker - * New special value COUNT which allows calculations based on the position of a - * value within a data set. Bug fix in rrd_rpncalc.c. PREV returned erroneus - * value for the second value. Bug fix in rrd_restore.c. Bug causing seek error - * when accesing an RRD restored from an xml that holds an RRD version <3. - * -- Ruben Justo - * - * Revision 1.7 2003/03/31 21:22:12 oetiker - * enables RRDtool updates with microsecond or in case of windows millisecond - * precision. This is needed to reduce time measurement error when archive step - * is small. (<30s) -- Sasha Mikheev - * - * Revision 1.6 2003/02/13 07:05:27 oetiker - * Find attached the patch I promised to send to you. Please note that there - * are three new source files (src/rrd_is_thread_safe.h, src/rrd_thread_safe.c - * and src/rrd_not_thread_safe.c) and the introduction of librrd_th. This - * library is identical to librrd, but it contains support code for per-thread - * global variables currently used for error information only. This is similar - * to how errno per-thread variables are implemented. librrd_th must be linked - * alongside of libpthred - * - * There is also a new file "THREADS", holding some documentation. - * - * -- Peter Stamfest - * - * Revision 1.5 2002/06/20 00:21:03 jake - * More Win32 build changes; thanks to Kerry Calvert. - * - * Revision 1.4 2002/02/01 20:34:49 oetiker - * fixed version number and date/time - * - * Revision 1.3 2001/03/04 13:01:55 oetiker - * Aberrant Behavior Detection support. A brief overview added to rrdtool.pod. - * Major updates to rrd_update.c, rrd_create.c. Minor update to other core files. - * This is backwards compatible! But new files using the Aberrant stuff are not readable - * by old rrdtool versions. See http://cricket.sourceforge.net/aberrant/rrd_hw.htm - * -- Jake Brutlag - * - * Revision 1.2 2001/03/04 10:29:20 oetiker - * fixed filedescriptor leak - * -- Mike Franusich - * - * Revision 1.1.1.1 2001/02/25 22:25:05 oetiker - * checkin - * *****************************************************************************/ #include "rrd_tool.h" #include "unused.h" + +#ifdef WIN32 +#include +#include +#include +#endif + + +#ifdef HAVE_BROKEN_MS_ASYNC +#include +#include +#endif + #define MEMBLK 8192 +#ifdef WIN32 +#define _LK_UNLCK 0 /* Unlock */ +#define _LK_LOCK 1 /* Lock */ +#define _LK_NBLCK 2 /* Non-blocking lock */ +#define _LK_RLCK 3 /* Lock for read only */ +#define _LK_NBRLCK 4 /* Non-blocking lock for read only */ + + +#define LK_UNLCK _LK_UNLCK +#define LK_LOCK _LK_LOCK +#define LK_NBLCK _LK_NBLCK +#define LK_RLCK _LK_RLCK +#define LK_NBRLCK _LK_NBRLCK +#endif + /* DEBUG 2 prints information obtained via mincore(2) */ -// #define DEBUG 2 -/* do not calculate exact madvise hints but assume 2 pages for headers and +#define DEBUG 1 +/* do not calculate exact madvise hints but assume 1 page for headers and * set DONTNEED for the rest, which is assumed to be data */ -//#define TWO_PAGES 1 /* Avoid calling madvise on areas that were already hinted. May be benefical if * your syscalls are very slow */ -//#define CHECK_MADVISE_OVERLAPS 1 #ifdef HAVE_MMAP -#define __rrd_read(dst, dst_t, cnt) \ - (dst) = (dst_t*) (data + offset); \ - offset += sizeof(dst_t) * (cnt) +/* the cast to void* is there to avoid this warning seen on ia64 with certain + versions of gcc: 'cast increases required alignment of target type' +*/ +#define __rrd_read(dst, dst_t, cnt) { \ + size_t wanted = sizeof(dst_t)*(cnt); \ + if (offset + wanted > rrd_file->file_len) { \ + rrd_set_error("reached EOF while loading header " #dst); \ + goto out_nullify_head; \ + } \ + (dst) = (dst_t*)(void*) (data + offset); \ + offset += wanted; \ + } #else -#define __rrd_read(dst, dst_t, cnt) \ - if ((dst = malloc(sizeof(dst_t)*(cnt))) == NULL) { \ +#define __rrd_read(dst, dst_t, cnt) { \ + size_t wanted = sizeof(dst_t)*(cnt); \ + size_t got; \ + if ((dst = (dst_t*)malloc(wanted)) == NULL) { \ rrd_set_error(#dst " malloc"); \ goto out_nullify_head; \ } \ - offset += read (rrd_file->fd, dst, sizeof(dst_t)*(cnt)) + got = read (rrd_simple_file->fd, dst, wanted); \ + if (got != wanted) { \ + rrd_set_error("short read while reading header " #dst); \ + goto out_nullify_head; \ + } \ + offset += got; \ + } #endif -/* next page-aligned (i.e. page-align up) */ -#ifndef PAGE_ALIGN -#define PAGE_ALIGN(addr) (((addr)+_page_size-1)&(~(_page_size-1))) +/* get the address of the start of this page */ +#if defined USE_MADVISE || defined HAVE_POSIX_FADVISE +#ifndef PAGE_START +#define PAGE_START(addr) ((addr)&(~(_page_size-1))) #endif -/* previous page-aligned (i.e. page-align down) */ -#ifndef PAGE_ALIGN_DOWN -#define PAGE_ALIGN_DOWN(addr) (((addr)+_page_size-1)&(~(_page_size-1))) #endif -#ifdef HAVE_MMAP -/* vector of last madvise hint */ -typedef struct _madvise_vec_t { - void *start; - ssize_t length; -} _madvise_vec_t; -_madvise_vec_t _madv_vec = { NULL, 0 }; -#endif - -#if defined CHECK_MADVISE_OVERLAPS -#define _madvise(_start, _off, _hint) \ - if ((_start) != _madv_vec.start && (ssize_t)(_off) != _madv_vec.length) { \ - _madv_vec.start = (_start) ; _madv_vec.length = (_off); \ - madvise((_start), (_off), (_hint)); \ - } -#else -#define _madvise(_start, _off, _hint) \ - madvise((_start), (_off), (_hint)) -#endif - -/* open a database file, return its header and an open filehandle */ -/* positioned to the first cdp in the first rra */ +/* Open a database file, return its header and an open filehandle, + * positioned to the first cdp in the first rra. + * In the error path of rrd_open, only rrd_free(&rrd) has to be called + * before returning an error. Do not call rrd_close upon failure of rrd_open. + * If creating a new file, the parameter rrd must be initialised with + * details of the file content. + * If opening an existing file, then use rrd must be initialised by + * rrd_init(rrd) prior to invoking rrd_open + */ rrd_file_t *rrd_open( const char *const file_name, rrd_t *rrd, unsigned rdwr) { + unsigned long ui; int flags = 0; - mode_t mode = S_IRUSR; int version; #ifdef HAVE_MMAP ssize_t _page_size = sysconf(_SC_PAGESIZE); - int mm_prot = PROT_READ, mm_flags = 0; - char *data; + char *data = MAP_FAILED; #endif off_t offset = 0; struct stat statb; - rrd_file_t *rrd_file = malloc(sizeof(rrd_file_t)); + rrd_file_t *rrd_file = NULL; + rrd_simple_file_t *rrd_simple_file = NULL; + size_t newfile_size = 0; + size_t header_len, value_cnt, data_len; + + /* Are we creating a new file? */ + if((rdwr & RRD_CREAT) && (rrd->stat_head != NULL)) + { + header_len = rrd_get_header_size(rrd); + + value_cnt = 0; + for (ui = 0; ui < rrd->stat_head->rra_cnt; ui++) + value_cnt += rrd->stat_head->ds_cnt * rrd->rra_def[ui].row_cnt; + data_len = sizeof(rrd_value_t) * value_cnt; + + newfile_size = header_len + data_len; + } + + rrd_file = (rrd_file_t*)malloc(sizeof(rrd_file_t)); if (rrd_file == NULL) { rrd_set_error("allocating rrd_file descriptor for '%s'", file_name); return NULL; } memset(rrd_file, 0, sizeof(rrd_file_t)); - rrd_init(rrd); + + rrd_file->pvt = malloc(sizeof(rrd_simple_file_t)); + if(rrd_file->pvt == NULL) { + rrd_set_error("allocating rrd_simple_file for '%s'", file_name); + return NULL; + } + memset(rrd_file->pvt, 0, sizeof(rrd_simple_file_t)); + rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt; + #ifdef DEBUG if ((rdwr & (RRD_READONLY | RRD_READWRITE)) == (RRD_READONLY | RRD_READWRITE)) { @@ -152,114 +149,135 @@ rrd_file_t *rrd_open( exit(-1); } #endif + +#ifdef HAVE_MMAP + rrd_simple_file->mm_prot = PROT_READ; + rrd_simple_file->mm_flags = 0; +#endif + if (rdwr & RRD_READONLY) { flags |= O_RDONLY; #ifdef HAVE_MMAP - mm_flags = MAP_PRIVATE; +# if !defined(AIX) + rrd_simple_file->mm_flags = MAP_PRIVATE; +# endif # ifdef MAP_NORESERVE - mm_flags |= MAP_NORESERVE; /* readonly, so no swap backing needed */ + rrd_simple_file->mm_flags |= MAP_NORESERVE; /* readonly, so no swap backing needed */ # endif #endif } else { if (rdwr & RRD_READWRITE) { - mode |= S_IWUSR; flags |= O_RDWR; -#ifdef HAVE_MMAP - mm_flags = MAP_SHARED; - mm_prot |= PROT_WRITE; -#endif +#ifdef HAVE_MMAP + rrd_simple_file->mm_flags = MAP_SHARED; + rrd_simple_file->mm_prot |= PROT_WRITE; +#endif } if (rdwr & RRD_CREAT) { flags |= (O_CREAT | O_TRUNC); } + if (rdwr & RRD_EXCL) { + flags |= O_EXCL; + } } if (rdwr & RRD_READAHEAD) { #ifdef MAP_POPULATE - mm_flags |= MAP_POPULATE; /* populate ptes and data */ + rrd_simple_file->mm_flags |= MAP_POPULATE; /* populate ptes and data */ #endif #if defined MAP_NONBLOCK - mm_flags |= MAP_NONBLOCK; /* just populate ptes */ -#endif -#ifdef USE_DIRECT_IO - } else { - flags |= O_DIRECT; + rrd_simple_file->mm_flags |= MAP_NONBLOCK; /* just populate ptes */ #endif } -#ifdef O_NONBLOCK - flags |= O_NONBLOCK; +#if defined(_WIN32) && !defined(__CYGWIN__) && !defined(__CYGWIN32__) + flags |= O_BINARY; #endif - if ((rrd_file->fd = open(file_name, flags, mode)) < 0) { + if ((rrd_simple_file->fd = open(file_name, flags, 0666)) < 0) { rrd_set_error("opening '%s': %s", file_name, rrd_strerror(errno)); - return NULL; + goto out_free; } +#ifdef HAVE_MMAP +#ifdef HAVE_BROKEN_MS_ASYNC + if (rdwr & RRD_READWRITE) { + /* some unices, the files mtime does not get update + on msync MS_ASYNC, in order to help them, + we update the the timestamp at this point. + The thing happens pretty 'close' to the open + call so the chances of a race should be minimal. + + Maybe ask your vendor to fix your OS ... */ + utime(file_name,NULL); + } +#endif +#endif + /* Better try to avoid seeks as much as possible. stat may be heavy but * many concurrent seeks are even worse. */ - if ((fstat(rrd_file->fd, &statb)) < 0) { + if (newfile_size == 0 && ((fstat(rrd_simple_file->fd, &statb)) < 0)) { rrd_set_error("fstat '%s': %s", file_name, rrd_strerror(errno)); goto out_close; } - rrd_file->file_len = statb.st_size; - + if (newfile_size == 0) { + rrd_file->file_len = statb.st_size; + } else { + rrd_file->file_len = newfile_size; + lseek(rrd_simple_file->fd, newfile_size - 1, SEEK_SET); + if ( write(rrd_simple_file->fd, "\0", 1) == -1){ /* poke */ + rrd_set_error("write '%s': %s", file_name, rrd_strerror(errno)); + goto out_close; + } + lseek(rrd_simple_file->fd, 0, SEEK_SET); + } #ifdef HAVE_POSIX_FADVISE /* In general we need no read-ahead when dealing with rrd_files. When we stop reading, it is highly unlikely that we start up again. In this manner we actually save time and diskaccess (and buffer cache). Thanks to Dave Plonka for the Idea of using POSIX_FADV_RANDOM here. */ - if (0 != posix_fadvise(rrd_file->fd, 0, 0, POSIX_FADV_RANDOM)) { - rrd_set_error("setting POSIX_FADV_RANDOM on '%s': %s", file_name, - rrd_strerror(errno)); - goto out_close; - } + posix_fadvise(rrd_simple_file->fd, 0, 0, POSIX_FADV_RANDOM); #endif /* if (rdwr & RRD_READWRITE) { - if (setvbuf((rrd_file->fd),NULL,_IONBF,2)) { + if (setvbuf((rrd_simple_file->fd),NULL,_IONBF,2)) { rrd_set_error("failed to disable the stream buffer\n"); return (-1); } } */ + #ifdef HAVE_MMAP - data = mmap(0, rrd_file->file_len, mm_prot, mm_flags, - rrd_file->fd, offset); + data = mmap(0, rrd_file->file_len, + rrd_simple_file->mm_prot, rrd_simple_file->mm_flags, + rrd_simple_file->fd, offset); /* lets see if the first read worked */ if (data == MAP_FAILED) { - rrd_set_error("error mmaping file '%s': %s", file_name, + rrd_set_error("mmaping file '%s': %s", file_name, rrd_strerror(errno)); goto out_close; } - rrd_file->file_start = data; + rrd_simple_file->file_start = data; + if (rdwr & RRD_CREAT) { + memset(data, DNAN, newfile_size - 1); + goto out_done; + } #endif + if (rdwr & RRD_CREAT) + goto out_done; #ifdef USE_MADVISE if (rdwr & RRD_COPY) { /* We will read everything in a moment (copying) */ - _madvise(data, rrd_file->file_len, MADV_WILLNEED | MADV_SEQUENTIAL); - goto out_done; + madvise(data, rrd_file->file_len, MADV_WILLNEED ); + madvise(data, rrd_file->file_len, MADV_SEQUENTIAL ); + } else { + /* We do not need to read anything in for the moment */ + madvise(data, rrd_file->file_len, MADV_RANDOM); + /* the stat_head will be needed soonish, so hint accordingly */ + madvise(data, sizeof(stat_head_t), MADV_WILLNEED); + madvise(data, sizeof(stat_head_t), MADV_RANDOM); } - /* We do not need to read anything in for the moment */ -#ifndef TWO_PAGES - _madvise(data, rrd_file->file_len, MADV_DONTNEED); -// _madvise(data, rrd_file->file_len, MADV_RANDOM); -#else -/* alternatively: keep 2 pages worth of data, likely headers, - * don't need the rest. */ - _madvise(data, _page_size * 2, MADV_WILLNEED | MADV_SEQUENTIAL); - _madvise(data + _page_size * 2, (rrd_file->file_len >= _page_size * 2) - ? rrd_file->file_len - _page_size * 2 : 0, MADV_DONTNEED); -#endif -#endif - -#if defined USE_MADVISE && !defined TWO_PAGES - /* the stat_head will be needed soonish, so hint accordingly */ -// too finegrained to calc the individual sizes, just keep 2 pages worth of hdr - _madvise(data + PAGE_ALIGN_DOWN(offset), PAGE_ALIGN(sizeof(stat_head_t)), - MADV_WILLNEED); - #endif __rrd_read(rrd->stat_head, stat_head_t, @@ -272,7 +290,7 @@ rrd_file_t *rrd_open( } if (rrd->stat_head->float_cookie != FLOAT_COOKIE) { - rrd_set_error("This RRD was created on other architecture"); + rrd_set_error("This RRD was created on another architecture"); goto out_nullify_head; } @@ -283,20 +301,18 @@ rrd_file_t *rrd_open( rrd->stat_head->version); goto out_nullify_head; } -#if defined USE_MADVISE && !defined TWO_PAGES +#if defined USE_MADVISE /* the ds_def will be needed soonish, so hint accordingly */ - _madvise(data + PAGE_ALIGN_DOWN(offset), - PAGE_ALIGN(sizeof(ds_def_t) * rrd->stat_head->ds_cnt), - MADV_WILLNEED); + madvise(data + PAGE_START(offset), + sizeof(ds_def_t) * rrd->stat_head->ds_cnt, MADV_WILLNEED); #endif __rrd_read(rrd->ds_def, ds_def_t, rrd->stat_head->ds_cnt); -#if defined USE_MADVISE && !defined TWO_PAGES +#if defined USE_MADVISE /* the rra_def will be needed soonish, so hint accordingly */ - _madvise(data + PAGE_ALIGN_DOWN(offset), - PAGE_ALIGN(sizeof(rra_def_t) * rrd->stat_head->rra_cnt), - MADV_WILLNEED); + madvise(data + PAGE_START(offset), + sizeof(rra_def_t) * rrd->stat_head->rra_cnt, MADV_WILLNEED); #endif __rrd_read(rrd->rra_def, rra_def_t, rrd->stat_head->rra_cnt); @@ -308,72 +324,90 @@ rrd_file_t *rrd_open( rrd_set_error("live_head_t malloc"); goto out_close; } -#ifdef HAVE_MMAP - memmove(&rrd->live_head->last_up, data + offset, sizeof(long)); - offset += sizeof(long); -#else - offset += read(rrd_file->fd, &rrd->live_head->last_up, sizeof(long)); +#if defined USE_MADVISE + /* the live_head will be needed soonish, so hint accordingly */ + madvise(data + PAGE_START(offset), sizeof(time_t), MADV_WILLNEED); #endif + __rrd_read(rrd->legacy_last_up, time_t, + 1); + + rrd->live_head->last_up = *rrd->legacy_last_up; rrd->live_head->last_up_usec = 0; } else { -#if defined USE_MADVISE && !defined TWO_PAGES +#if defined USE_MADVISE /* the live_head will be needed soonish, so hint accordingly */ - _madvise(data + PAGE_ALIGN_DOWN(offset), - PAGE_ALIGN(sizeof(live_head_t)), MADV_WILLNEED); + madvise(data + PAGE_START(offset), + sizeof(live_head_t), MADV_WILLNEED); #endif __rrd_read(rrd->live_head, live_head_t, 1); } -//XXX: This doesn't look like it needs madvise __rrd_read(rrd->pdp_prep, pdp_prep_t, rrd->stat_head->ds_cnt); - -//XXX: This could benefit from madvise()ing __rrd_read(rrd->cdp_prep, cdp_prep_t, rrd->stat_head->rra_cnt * rrd->stat_head->ds_cnt); - -//XXX: This could benefit from madvise()ing __rrd_read(rrd->rra_ptr, rra_ptr_t, rrd->stat_head->rra_cnt); -#ifdef USE_MADVISE - out_done: -#endif rrd_file->header_len = offset; rrd_file->pos = offset; + { + unsigned long row_cnt = 0; + + for (ui=0; uistat_head->rra_cnt; ui++) + row_cnt += rrd->rra_def[ui].row_cnt; + + size_t correct_len = rrd_file->header_len + + sizeof(rrd_value_t) * row_cnt * rrd->stat_head->ds_cnt; + + if (correct_len > rrd_file->file_len) + { + rrd_set_error("'%s' is too small (should be %ld bytes)", + file_name, (long long) correct_len); + goto out_nullify_head; + } + } + + out_done: return (rrd_file); out_nullify_head: rrd->stat_head = NULL; out_close: - close(rrd_file->fd); +#ifdef HAVE_MMAP + if (data != MAP_FAILED) + munmap(data, rrd_file->file_len); +#endif + + close(rrd_simple_file->fd); + out_free: + free(rrd_file->pvt); + free(rrd_file); return NULL; } -/* Close a reference to an rrd_file. */ - -int rrd_close( - rrd_file_t *rrd_file) -{ - int ret; - -#if defined HAVE_MMAP - ssize_t _page_size = sysconf(_SC_PAGESIZE); -#endif #if defined DEBUG && DEBUG > 1 +/* Print list of in-core pages of a the current rrd_file. */ +static +void mincore_print( + rrd_file_t *rrd_file, + char *mark) +{ + rrd_simple_file_t *rrd_simple_file; + rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt; +#ifdef HAVE_MMAP /* pretty print blocks in core */ - off_t off; + size_t off; unsigned char *vec; + ssize_t _page_size = sysconf(_SC_PAGESIZE); - off = - rrd_file->file_len + - ((rrd_file->file_len + sysconf(_SC_PAGESIZE) - - 1) / sysconf(_SC_PAGESIZE)); + off = rrd_file->file_len + + ((rrd_file->file_len + _page_size - 1) / _page_size); vec = malloc(off); if (vec != NULL) { memset(vec, 0, off); - if (mincore(rrd_file->file_start, rrd_file->file_len, vec) == 0) { + if (mincore(rrd_simple_file->file_start, rrd_file->file_len, vec) == 0) { int prev; unsigned is_in = 0, was_in = 0; @@ -382,46 +416,160 @@ int rrd_close( if (off == 0) was_in = is_in; if (was_in != is_in) { - fprintf(stderr, "%sin core: %p len %ld\n", + fprintf(stderr, "%s: %sin core: %p len %ld\n", mark, was_in ? "" : "not ", vec + prev, off - prev); was_in = is_in; prev = off; } } fprintf(stderr, - "%sin core: %p len %ld\n", + "%s: %sin core: %p len %ld\n", mark, was_in ? "" : "not ", vec + prev, off - prev); } else fprintf(stderr, "mincore: %s", rrd_strerror(errno)); } -#endif /* DEBUG */ +#else + fprintf(stderr, "sorry mincore only works with mmap"); +#endif +} +#endif /* defined DEBUG && DEBUG > 1 */ -#ifdef USE_MADVISE -#ifdef TWO_PAGES -//XXX: ? - /* Keep 2 pages worth of headers around, round up to next page boundary. */ - ret = - PAGE_ALIGN(rrd_file->header_len % _page_size + rrd_file->header_len); - if (rrd_file->file_len > ret) - _madvise(rrd_file->file_start + ret, - rrd_file->file_len - ret, MADV_DONTNEED); +/* + * get exclusive lock to whole file. + * lock gets removed when we close the file + * + * returns 0 on success + */ +int rrd_lock( + rrd_file_t *rrd_file) +{ + int rcstat; + rrd_simple_file_t *rrd_simple_file; + rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt; + + { +#if defined(_WIN32) && !defined(__CYGWIN__) && !defined(__CYGWIN32__) + struct _stat st; + + if (_fstat(rrd_simple_file->fd, &st) == 0) { + rcstat = _locking(rrd_simple_file->fd, _LK_NBLCK, st.st_size); + } else { + rcstat = -1; + } #else + struct flock lock; + + lock.l_type = F_WRLCK; /* exclusive write lock */ + lock.l_len = 0; /* whole file */ + lock.l_start = 0; /* start of file */ + lock.l_whence = SEEK_SET; /* end of file */ + + rcstat = fcntl(rrd_simple_file->fd, F_SETLK, &lock); +#endif + } + + return (rcstat); +} + + +/* drop cache except for the header and the active pages */ +void rrd_dontneed( + rrd_file_t *rrd_file, + rrd_t *rrd) +{ + rrd_simple_file_t *rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt; +#if defined USE_MADVISE || defined HAVE_POSIX_FADVISE + size_t dontneed_start; + size_t rra_start; + size_t active_block; + size_t i; + ssize_t _page_size = sysconf(_SC_PAGESIZE); + + if (rrd_file == NULL) { +#if defined DEBUG && DEBUG + fprintf (stderr, "rrd_dontneed: Argument 'rrd_file' is NULL.\n"); +#endif + return; + } + +#if defined DEBUG && DEBUG > 1 + mincore_print(rrd_file, "before"); +#endif + /* ignoring errors from RRDs that are smaller then the file_len+rounding */ - _madvise(rrd_file->file_start + PAGE_ALIGN_DOWN(rrd_file->header_len), - rrd_file->file_len - PAGE_ALIGN(rrd_file->header_len), - MADV_DONTNEED); + rra_start = rrd_file->header_len; + dontneed_start = PAGE_START(rra_start) + _page_size; + for (i = 0; i < rrd->stat_head->rra_cnt; ++i) { + active_block = + PAGE_START(rra_start + + rrd->rra_ptr[i].cur_row + * rrd->stat_head->ds_cnt * sizeof(rrd_value_t)); + if (active_block > dontneed_start) { +#ifdef USE_MADVISE + madvise(rrd_simple_file->file_start + dontneed_start, + active_block - dontneed_start - 1, MADV_DONTNEED); +#endif +/* in linux at least only fadvise DONTNEED seems to purge pages from cache */ +#ifdef HAVE_POSIX_FADVISE + posix_fadvise(rrd_simple_file->fd, dontneed_start, + active_block - dontneed_start - 1, + POSIX_FADV_DONTNEED); +#endif + } + dontneed_start = active_block; + /* do not release 'hot' block if update for this RAA will occur + * within 10 minutes */ + if (rrd->stat_head->pdp_step * rrd->rra_def[i].pdp_cnt - + rrd->live_head->last_up % (rrd->stat_head->pdp_step * + rrd->rra_def[i].pdp_cnt) < 10 * 60) { + dontneed_start += _page_size; + } + rra_start += + rrd->rra_def[i].row_cnt * rrd->stat_head->ds_cnt * + sizeof(rrd_value_t); + } + + if (dontneed_start < rrd_file->file_len) { +#ifdef USE_MADVISE + madvise(rrd_simple_file->file_start + dontneed_start, + rrd_file->file_len - dontneed_start, MADV_DONTNEED); +#endif +#ifdef HAVE_POSIX_FADVISE + posix_fadvise(rrd_simple_file->fd, dontneed_start, + rrd_file->file_len - dontneed_start, + POSIX_FADV_DONTNEED); #endif + } + +#if defined DEBUG && DEBUG > 1 + mincore_print(rrd_file, "after"); #endif +#endif /* without madvise and posix_fadvise it does not make much sense todo anything */ +} + + + + + +int rrd_close( + rrd_file_t *rrd_file) +{ + rrd_simple_file_t *rrd_simple_file; + rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt; + int ret; + #ifdef HAVE_MMAP - ret = munmap(rrd_file->file_start, rrd_file->file_len); + ret = msync(rrd_simple_file->file_start, rrd_file->file_len, MS_ASYNC); + if (ret != 0) + rrd_set_error("msync rrd_file: %s", rrd_strerror(errno)); + ret = munmap(rrd_simple_file->file_start, rrd_file->file_len); if (ret != 0) rrd_set_error("munmap rrd_file: %s", rrd_strerror(errno)); -#else - ret = 0; #endif -// ret = close(rrd_file->fd); -// if (ret != 0) -// rrd_set_error("closing file: %s", rrd_strerror(errno)); + ret = close(rrd_simple_file->fd); + if (ret != 0) + rrd_set_error("closing file: %s", rrd_strerror(errno)); + free(rrd_file->pvt); free(rrd_file); rrd_file = NULL; return ret; @@ -436,6 +584,8 @@ off_t rrd_seek( int whence) { off_t ret = 0; + rrd_simple_file_t *rrd_simple_file; + rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt; #ifdef HAVE_MMAP if (whence == SEEK_SET) @@ -445,77 +595,105 @@ off_t rrd_seek( else if (whence == SEEK_END) rrd_file->pos = rrd_file->file_len + off; #else - ret = lseek(rrd_file->fd, off, whence); + ret = lseek(rrd_simple_file->fd, off, whence); if (ret < 0) rrd_set_error("lseek: %s", rrd_strerror(errno)); rrd_file->pos = ret; #endif -//XXX: mimic fseek, which returns 0 upon success - return ret == -1; //XXX: or just ret to mimic lseek + /* mimic fseek, which returns 0 upon success */ + return ret < 0; /*XXX: or just ret to mimic lseek */ } /* Get current position in rrd_file. */ -inline off_t rrd_tell(rrd_file_t *rrd_file) +off_t rrd_tell( + rrd_file_t *rrd_file) { return rrd_file->pos; } -/* read count bytes into buffer buf, starting at rrd_file->pos. - * Returns the number of bytes read. */ +/* Read count bytes into buffer buf, starting at rrd_file->pos. + * Returns the number of bytes read or <0 on error. */ ssize_t rrd_read( rrd_file_t *rrd_file, void *buf, size_t count) { + rrd_simple_file_t *rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt; #ifdef HAVE_MMAP - buf = memmove(buf, rrd_file->file_start + rrd_file->pos, count); - rrd_file->pos += count; /* mimmic read() semantics */ - return count; + size_t _cnt = count; + ssize_t _surplus; + + if (rrd_file->pos > rrd_file->file_len || _cnt == 0) /* EOF */ + return 0; + if (buf == NULL) + return -1; /* EINVAL */ + _surplus = rrd_file->pos + _cnt - rrd_file->file_len; + if (_surplus > 0) { /* short read */ + _cnt -= _surplus; + } + if (_cnt == 0) + return 0; /* EOF */ + buf = memcpy(buf, rrd_simple_file->file_start + rrd_file->pos, _cnt); + + rrd_file->pos += _cnt; /* mimmic read() semantics */ + return _cnt; #else ssize_t ret; - ret = read(rrd_file->fd, buf, count); - //XXX: eventually add generic rrd_set_error(""); here - rrd_file->pos += count; /* mimmic read() semantics */ + ret = read(rrd_simple_file->fd, buf, count); + if (ret > 0) + rrd_file->pos += ret; /* mimmic read() semantics */ return ret; #endif } -/* write count bytes from buffer buf to the current position - * rrd_file->pos of rrd_file->fd. - * Returns the number of bytes written. */ +/* Write count bytes from buffer buf to the current position + * rrd_file->pos of rrd_simple_file->fd. + * Returns the number of bytes written or <0 on error. */ ssize_t rrd_write( rrd_file_t *rrd_file, const void *buf, size_t count) { + rrd_simple_file_t *rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt; #ifdef HAVE_MMAP - memmove(rrd_file->file_start + rrd_file->pos, buf, count); + size_t old_size = rrd_file->file_len; + if (count == 0) + return 0; + if (buf == NULL) + return -1; /* EINVAL */ + + if((rrd_file->pos + count) > old_size) + { + rrd_set_error("attempting to write beyond end of file"); + return -1; + } + memcpy(rrd_simple_file->file_start + rrd_file->pos, buf, count); + rrd_file->pos += count; return count; /* mimmic write() semantics */ #else - return write(rrd_file->fd, buf, count); + ssize_t _sz = write(rrd_simple_file->fd, buf, count); + + if (_sz > 0) + rrd_file->pos += _sz; + return _sz; #endif } -/* flush all data pending to be written to FD. */ - -inline void rrd_flush( - rrd_file_t *rrd_file) +/* this is a leftover from the old days, it serves no purpose + and is therefore turned into a no-op */ +void rrd_flush( + rrd_file_t UNUSED(*rrd_file)) { - if (fdatasync(rrd_file->fd) != 0) { - rrd_set_error("flushing fd %d: %s", rrd_file->fd, - rrd_strerror(errno)); - } } - /* Initialize RRD header. */ void rrd_init( @@ -525,6 +703,7 @@ void rrd_init( rrd->ds_def = NULL; rrd->rra_def = NULL; rrd->live_head = NULL; + rrd->legacy_last_up = NULL; rrd->rra_ptr = NULL; rrd->pdp_prep = NULL; rrd->cdp_prep = NULL; @@ -534,12 +713,19 @@ void rrd_init( /* free RRD header data. */ +#ifdef HAVE_MMAP void rrd_free( - rrd_t UNUSED(*rrd)) + rrd_t *rrd) { -#ifndef HAVE_MMAP - if (atoi(rrd->stat_head->version) < 3) + if (rrd->legacy_last_up) { /* this gets set for version < 3 only */ free(rrd->live_head); + } +} +#else +void rrd_free( + rrd_t *rrd) +{ + free(rrd->live_head); free(rrd->stat_head); free(rrd->ds_def); free(rrd->rra_def); @@ -547,8 +733,8 @@ void rrd_free( free(rrd->pdp_prep); free(rrd->cdp_prep); free(rrd->rrd_value); -#endif } +#endif /* routine used by external libraries to free memory allocated by @@ -560,63 +746,31 @@ void rrd_freemem( free(mem); } - -/* XXX: FIXME: missing documentation. */ -/*XXX: FIXME should be renamed to rrd_readfile or _rrd_readfile */ - -int /*_rrd_*/ readfile( - const char *file_name, - char **buffer, - int skipfirst) +/* + * rra_update informs us about the RRAs being updated + * The low level storage API may use this information for + * aligning RRAs within stripes, or other performance enhancements + */ +void rrd_notify_row( + rrd_file_t UNUSED(*rrd_file), + int UNUSED(rra_idx), + unsigned long UNUSED(rra_row), + time_t UNUSED(rra_time)) { - long writecnt = 0, totalcnt = MEMBLK; - long offset = 0; - FILE *input = NULL; - char c; +} - if ((strcmp("-", file_name) == 0)) { - input = stdin; - } else { - if ((input = fopen(file_name, "rb")) == NULL) { - rrd_set_error("opening '%s': %s", file_name, rrd_strerror(errno)); - return (-1); - } - } - if (skipfirst) { - do { - c = getc(input); - offset++; - } while (c != '\n' && !feof(input)); - } - if (strcmp("-", file_name)) { - fseek(input, 0, SEEK_END); - /* have extra space for detecting EOF without realloc */ - totalcnt = (ftell(input) + 1) / sizeof(char) - offset; - if (totalcnt < MEMBLK) - totalcnt = MEMBLK; /* sanitize */ - fseek(input, offset * sizeof(char), SEEK_SET); - } - if (((*buffer) = (char *) malloc((totalcnt + 4) * sizeof(char))) == NULL) { - perror("Allocate Buffer:"); - exit(1); - }; - do { - writecnt += - fread((*buffer) + writecnt, 1, - (totalcnt - writecnt) * sizeof(char), input); - if (writecnt >= totalcnt) { - totalcnt += MEMBLK; - if (((*buffer) = - rrd_realloc((*buffer), - (totalcnt + 4) * sizeof(char))) == NULL) { - perror("Realloc Buffer:"); - exit(1); - }; - } - } while (!feof(input)); - (*buffer)[writecnt] = '\0'; - if (strcmp("-", file_name) != 0) { - fclose(input); - }; - return writecnt; +/* + * This function is called when creating a new RRD + * The storage implementation can use this opportunity to select + * a sensible starting row within the file. + * The default implementation is random, to ensure that all RRAs + * don't change to a new disk block at the same time + */ +unsigned long rrd_select_initial_row( + rrd_file_t UNUSED(*rrd_file), + int UNUSED(rra_idx), + rra_def_t *rra + ) +{ + return rrd_random() % rra->row_cnt; }