track rsync-fadvise.patch locally (in this repo)

[packages/rsync.git] / rsync-fadvise.patch

diff -ru rsync-3.1.2.orig/checksum.c rsync-3.1.2/checksum.c
--- rsync-3.1.2.orig/checksum.c	2015-08-08 22:47:03.000000000 +0300
+++ rsync-3.1.2/checksum.c	2016-10-24 15:38:28.002415712 +0300
@@ -24,6 +24,9 @@
 extern int checksum_seed;
 extern int protocol_version;
 extern int proper_seed_order;
+#ifdef WITH_DROP_CACHE
+#define close(fd) fadv_close(fd)
+#endif
 
 /*
   a simple 32 bit checksum that can be upadted from either end
diff -ru rsync-3.1.2.orig/cleanup.c rsync-3.1.2/cleanup.c
--- rsync-3.1.2.orig/cleanup.c	2015-08-08 22:47:03.000000000 +0300
+++ rsync-3.1.2/cleanup.c	2016-10-24 15:38:28.002415712 +0300
@@ -53,7 +53,11 @@
 	int fd;
 	int ret;
 	STRUCT_STAT st;
-
+#endif
+#ifdef WITH_DROP_CACHE
+	fadv_close_all();
+#endif
+#ifdef SHUTDOWN_ALL_SOCKETS
 	max_fd = sysconf(_SC_OPEN_MAX) - 1;
 	for (fd = max_fd; fd >= 0; fd--) {
 		if ((ret = do_fstat(fd, &st)) == 0) {
diff -ru rsync-3.1.2.orig/config.h.in rsync-3.1.2/config.h.in
--- rsync-3.1.2.orig/config.h.in	2015-12-21 22:20:53.000000000 +0200
+++ rsync-3.1.2/config.h.in	2016-10-24 15:38:28.006415712 +0300
@@ -275,6 +275,9 @@
 /* Define to 1 if you have the <memory.h> header file. */
 #undef HAVE_MEMORY_H
 
+/* Define to 1 if you have the `mincore' function. */
+#undef HAVE_MINCORE
+
 /* Define to 1 if you have the `mkfifo' function. */
 #undef HAVE_MKFIFO
 
@@ -287,6 +290,9 @@
 /* Define to 1 if the system has the type `mode_t'. */
 #undef HAVE_MODE_T
 
+/* Define to 1 if you have the `mmap' function. */
+#undef HAVE_MMAP
+
 /* Define to 1 if you have the `mtrace' function. */
 #undef HAVE_MTRACE
 
@@ -329,6 +335,9 @@
 /* true if you have posix ACLs */
 #undef HAVE_POSIX_ACLS
 
+/* Define to 1 if you have the `posix_fadvise64' function. */
+#undef HAVE_POSIX_FADVISE64
+
 /* Define to 1 if you have the `posix_fallocate' function. */
 #undef HAVE_POSIX_FALLOCATE
 
diff -ru rsync-3.1.2.orig/configure.ac rsync-3.1.2/configure.ac
--- rsync-3.1.2.orig/configure.ac	2015-12-21 22:00:49.000000000 +0200
+++ rsync-3.1.2/configure.ac	2016-10-24 15:38:28.006415712 +0300
@@ -598,6 +598,7 @@
     setlocale setmode open64 lseek64 mkstemp64 mtrace va_copy __va_copy \
     seteuid strerror putenv iconv_open locale_charset nl_langinfo getxattr \
     extattr_get_link sigaction sigprocmask setattrlist getgrouplist \
+    mmap mincore posix_fadvise64 \
     initgroups utimensat posix_fallocate attropen setvbuf usleep)
 
 dnl cygwin iconv.h defines iconv_open as libiconv_open
diff -ru rsync-3.1.2.orig/configure.sh rsync-3.1.2/configure.sh
--- rsync-3.1.2.orig/configure.sh	2015-12-21 22:20:53.000000000 +0200
+++ rsync-3.1.2/configure.sh	2016-10-24 15:38:28.006415712 +0300
@@ -7692,6 +7692,7 @@
     setlocale setmode open64 lseek64 mkstemp64 mtrace va_copy __va_copy \
     seteuid strerror putenv iconv_open locale_charset nl_langinfo getxattr \
     extattr_get_link sigaction sigprocmask setattrlist getgrouplist \
+    mmap mincore posix_fadvise64 \
     initgroups utimensat posix_fallocate attropen setvbuf usleep
 do :
   as_ac_var=`$as_echo "ac_cv_func_$ac_func" | $as_tr_sh`
diff -ru rsync-3.1.2.orig/fileio.c rsync-3.1.2/fileio.c
--- rsync-3.1.2.orig/fileio.c	2015-08-08 22:47:03.000000000 +0300
+++ rsync-3.1.2/fileio.c	2016-10-24 15:38:28.006415712 +0300
@@ -51,7 +51,7 @@
 		ret = -1;
 	else {
 		do {
-			ret = write(f, "", 1);
+			ret = fadv_write(f, "", 1);
 		} while (ret < 0 && errno == EINTR);
 
 		ret = ret <= 0 ? -1 : 0;
@@ -81,7 +81,7 @@
 		do_lseek(f, sparse_seek, SEEK_CUR);
 	sparse_seek = l2;
 
-	while ((ret = write(f, buf + l1, len - (l1+l2))) <= 0) {
+	while ((ret = fadv_write(f, buf + l1, len - (l1+l2))) <= 0) {
 		if (ret < 0 && errno == EINTR)
 			continue;
 		sparse_seek = 0;
@@ -107,7 +107,7 @@
 	char *bp = wf_writeBuf;
 
 	while (wf_writeBufCnt > 0) {
-		if ((ret = write(f, bp, wf_writeBufCnt)) < 0) {
+		if ((ret = fadv_write(f, bp, wf_writeBufCnt)) < 0) {
 			if (errno == EINTR)
 				continue;
 			return ret;
@@ -254,7 +254,7 @@
 	map->p_len = window_size;
 
 	while (read_size > 0) {
-		int32 nread = read(map->fd, map->p + read_offset, read_size);
+		int32 nread = fadv_read(map->fd, map->p + read_offset, read_size);
 		if (nread <= 0) {
 			if (!map->status)
 				map->status = nread ? errno : ENODATA;
diff -ru rsync-3.1.2.orig/generator.c rsync-3.1.2/generator.c
--- rsync-3.1.2.orig/generator.c	2015-12-05 21:10:24.000000000 +0200
+++ rsync-3.1.2/generator.c	2016-10-24 15:38:28.006415712 +0300
@@ -111,6 +111,10 @@
 static int need_retouch_dir_perms;
 static const char *solo_file = NULL;
 
+#ifdef WITH_DROP_CACHE
+#define close(fd) fadv_close(fd)
+#endif
+
 enum nonregtype {
     TYPE_DIR, TYPE_SPECIAL, TYPE_DEVICE, TYPE_SYMLINK
 };
diff -ru rsync-3.1.2.orig/options.c rsync-3.1.2/options.c
--- rsync-3.1.2.orig/options.c	2015-12-19 00:46:28.000000000 +0200
+++ rsync-3.1.2/options.c	2016-10-24 15:38:28.006415712 +0300
@@ -62,6 +62,9 @@
 int preserve_gid = 0;
 int preserve_times = 0;
 int update_only = 0;
+#ifdef WITH_DROP_CACHE
+int drop_cache = 0;
+#endif
 int cvs_exclude = 0;
 int dry_run = 0;
 int do_xfers = 1;
@@ -680,6 +683,9 @@
   rprintf(F,"     --backup-dir=DIR        make backups into hierarchy based in DIR\n");
   rprintf(F,"     --suffix=SUFFIX         set backup suffix (default %s w/o --backup-dir)\n",BACKUP_SUFFIX);
   rprintf(F," -u, --update                skip files that are newer on the receiver\n");
+#ifdef WITH_DROP_CACHE
+  rprintf(F,"     --drop-cache            do not cache rsync files (POSIX_FADV_DONTNEED)\n");
+#endif
   rprintf(F,"     --inplace               update destination files in-place (SEE MAN PAGE)\n");
   rprintf(F,"     --append                append data onto shorter files\n");
   rprintf(F,"     --append-verify         like --append, but with old data in file checksum\n");
@@ -914,6 +920,9 @@
   {"no-one-file-system",0, POPT_ARG_VAL,    &one_file_system, 0, 0, 0 },
   {"no-x",             0,  POPT_ARG_VAL,    &one_file_system, 0, 0, 0 },
   {"update",          'u', POPT_ARG_NONE,   &update_only, 0, 0, 0 },
+#ifdef WITH_DROP_CACHE
+  {"drop-cache",       0,  POPT_ARG_NONE,   &drop_cache, 0, 0, 0 },
+#endif
   {"existing",         0,  POPT_ARG_NONE,   &ignore_non_existing, 0, 0, 0 },
   {"ignore-non-existing",0,POPT_ARG_NONE,   &ignore_non_existing, 0, 0, 0 },
   {"ignore-existing",  0,  POPT_ARG_NONE,   &ignore_existing, 0, 0, 0 },
@@ -1065,6 +1074,9 @@
   rprintf(F,"     --log-file=FILE         override the \"log file\" setting\n");
   rprintf(F,"     --log-file-format=FMT   override the \"log format\" setting\n");
   rprintf(F,"     --sockopts=OPTIONS      specify custom TCP options\n");
+#ifdef WITH_DROP_CACHE
+  rprintf(F,"     --drop-cache            do not cache rsync files (POSIX_FADV_DONTNEED)\n");
+#endif
   rprintf(F," -v, --verbose               increase verbosity\n");
   rprintf(F," -4, --ipv4                  prefer IPv4\n");
   rprintf(F," -6, --ipv6                  prefer IPv6\n");
@@ -1089,6 +1101,9 @@
   {"log-file",         0,  POPT_ARG_STRING, &logfile_name, 0, 0, 0 },
   {"log-file-format",  0,  POPT_ARG_STRING, &logfile_format, 0, 0, 0 },
   {"port",             0,  POPT_ARG_INT,    &rsync_port, 0, 0, 0 },
+#ifdef WITH_DROP_CACHE
+  {"drop-cache",       0,  POPT_ARG_NONE,   &drop_cache, 0, 0, 0 },
+#endif
   {"sockopts",         0,  POPT_ARG_STRING, &sockopts, 0, 0, 0 },
   {"protocol",         0,  POPT_ARG_INT,    &protocol_version, 0, 0, 0 },
   {"server",           0,  POPT_ARG_NONE,   &am_server, 0, 0, 0 },
@@ -2376,6 +2391,11 @@
 	if (!am_sender)
 		args[ac++] = "--sender";
 
+#ifdef WITH_DROP_CACHE
+	if (drop_cache)
+		args[ac++] = "--drop-cache";
+#endif
+
 	x = 1;
 	argstr[0] = '-';
 
diff -ru rsync-3.1.2.orig/proto.h rsync-3.1.2/proto.h
--- rsync-3.1.2.orig/proto.h	2015-12-21 22:22:53.000000000 +0200
+++ rsync-3.1.2/proto.h	2016-10-24 15:38:28.010415712 +0300
@@ -348,6 +348,10 @@
 uid_t recv_user_name(int f, uid_t uid);
 gid_t recv_group_name(int f, gid_t gid, uint16 *flags_ptr);
 void recv_id_list(int f, struct file_list *flist);
+ssize_t fadv_write(int fd, const void *buf, size_t count);
+ssize_t fadv_read(int fd, void *buf, size_t count);
+void fadv_close_all(void);
+int fadv_close(int fd);
 void parse_name_map(char *map, BOOL usernames);
 const char *getallgroups(uid_t uid, item_list *gid_list);
 void set_nonblocking(int fd);
diff -ru rsync-3.1.2.orig/receiver.c rsync-3.1.2/receiver.c
--- rsync-3.1.2.orig/receiver.c	2015-09-07 20:07:17.000000000 +0300
+++ rsync-3.1.2/receiver.c	2016-10-24 15:38:28.010415712 +0300
@@ -62,6 +62,10 @@
 extern struct file_list *cur_flist, *first_flist, *dir_flist;
 extern filter_rule_list daemon_filter_list;
 
+#ifdef WITH_DROP_CACHE
+#define close(fd) fadv_close(fd)
+#endif
+
 static struct bitbag *delayed_bits = NULL;
 static int phase = 0, redoing = 0;
 static flist_ndx_list batch_redo_list;
diff -ru rsync-3.1.2.orig/rsync.1 rsync-3.1.2/rsync.1
--- rsync-3.1.2.orig/rsync.1	2015-12-21 22:22:41.000000000 +0200
+++ rsync-3.1.2/rsync.1	2016-10-24 15:38:28.010415712 +0300
@@ -453,6 +453,7 @@
      \-\-super                 receiver attempts super\-user activities
      \-\-fake\-super            store/recover privileged attrs using xattrs
  \-S, \-\-sparse                handle sparse files efficiently
+     \-\-drop\-cache            drop cache continuosly using fadvise
      \-\-preallocate           allocate dest files before writing
  \-n, \-\-dry\-run               perform a trial run with no changes made
  \-W, \-\-whole\-file            copy files whole (w/o delta\-xfer algorithm)
@@ -1426,6 +1427,13 @@
 up less space on the destination.  Conflicts with \fB\-\-inplace\fP because it\(cq\&s
 not possible to overwrite data in a sparse fashion.
 .IP 
+.IP "\fB\-\-drop\-cache\fP"
+Stop rsync from filling up the file system cache with the files it copies\&. Without this
+option other processes, that had been crunching along happily on your system, will suddenly
+become slow as they find their data being outsed from the cache. The \fB\-\-drop\-cache\fP function
+uses posix_fadvise64 and mincore todo its work\&. It will only get compiled if configure can find posix_fadvise64 and mincore\&.
+Rsync will tries only to drop data from cache that has not been cached before.
+.IP 
 .IP "\fB\-\-preallocate\fP"
 This tells the receiver to allocate each destination
 file to its eventual size before writing data to the file.  Rsync will only use
diff -ru rsync-3.1.2.orig/rsync.h rsync-3.1.2/rsync.h
--- rsync-3.1.2.orig/rsync.h	2015-08-08 22:47:03.000000000 +0300
+++ rsync-3.1.2/rsync.h	2016-10-24 15:38:28.010415712 +0300
@@ -1291,3 +1291,13 @@
 #ifdef MAINTAINER_MODE
 const char *get_panic_action(void);
 #endif
+
+#if defined HAVE_POSIX_FADVISE64 && defined HAVE_MINCORE && defined HAVE_MMAP
+#define WITH_DROP_CACHE 1
+#include <sys/mman.h>
+int fadv_close(int fd);
+void fadv_close_all(void);
+#endif
+
+ssize_t fadv_write(int fd, const void *buf, size_t count);
+ssize_t fadv_read(int fd, void *buf, size_t count);
diff -ru rsync-3.1.2.orig/rsync.yo rsync-3.1.2/rsync.yo
--- rsync-3.1.2.orig/rsync.yo	2015-12-21 22:00:49.000000000 +0200
+++ rsync-3.1.2/rsync.yo	2016-10-24 15:38:28.010415712 +0300
@@ -1244,6 +1244,17 @@
 up less space on the destination.  Conflicts with bf(--inplace) because it's
 not possible to overwrite data in a sparse fashion.
 
+dit(bf(--drop-cache)) Stop rsync from disturbing the file system cache with
+the data from the files it copies. Without this option other processes, that
+had been crunching along happily using cached data, will suddenly become
+slow as they find their favorite data blocks data being evicted from the
+cache by the files read and written by rsync. Since rsync has to wait until
+the data is written to disk, before it can drop the cache, this option will
+slow rsync down considerably, especially with small files and short copy
+jobs. The bf(--drop-cache) function uses posix_fadvise64 and mincore todo
+its work. It will only get compiled if configure can find posix_fadvise64
+and mincore.
+
 dit(bf(--preallocate)) This tells the receiver to allocate each destination
 file to its eventual size before writing data to the file.  Rsync will only use
 the real filesystem-level preallocation support provided by Linux's
diff -ru rsync-3.1.2.orig/sender.c rsync-3.1.2/sender.c
--- rsync-3.1.2.orig/sender.c	2015-09-07 20:07:17.000000000 +0300
+++ rsync-3.1.2/sender.c	2016-10-24 15:38:28.010415712 +0300
@@ -46,6 +46,9 @@
 extern int file_old_total;
 extern struct stats stats;
 extern struct file_list *cur_flist, *first_flist, *dir_flist;
+#ifdef WITH_DROP_CACHE
+#define close(fd) fadv_close(fd)
+#endif
 
 BOOL extra_flist_sending_enabled;
 
diff -ru rsync-3.1.2.orig/t_unsafe.c rsync-3.1.2/t_unsafe.c
--- rsync-3.1.2.orig/t_unsafe.c	2015-08-08 22:47:03.000000000 +0300
+++ rsync-3.1.2/t_unsafe.c	2016-10-24 15:38:28.010415712 +0300
@@ -24,6 +24,7 @@
 #include "rsync.h"
 
 int dry_run = 0;
+int drop_cache = 0;
 int am_root = 0;
 int am_sender = 1;
 int read_only = 0;
diff -ru rsync-3.1.2.orig/util.c rsync-3.1.2/util.c
--- rsync-3.1.2.orig/util.c	2015-12-21 20:54:02.000000000 +0200
+++ rsync-3.1.2/util.c	2016-10-24 15:38:28.014415712 +0300
@@ -37,6 +37,10 @@
 extern unsigned int module_dirlen;
 extern char *partial_dir;
 extern filter_rule_list daemon_filter_list;
+#ifdef WITH_DROP_CACHE
+#include <sys/mman.h>
+extern int drop_cache;
+#endif
 
 int sanitize_paths = 0;
 
@@ -44,6 +48,218 @@
 unsigned int curr_dir_len;
 int curr_dir_depth; /* This is only set for a sanitizing daemon. */
 
+#ifdef WITH_DROP_CACHE
+#define FADV_BUFFER_SIZE  1024*1024*16
+
+static struct stat fadv_fd_stat[1024];
+static off_t fadv_fd_pos[1024];
+static unsigned char *fadv_core_ptr[1024];
+static int fadv_max_fd = 0;
+static int fadv_close_ring_tail = 0;
+static int fadv_close_ring_head = 0;
+static int fadv_close_ring_size = 0;
+static int fadv_close_ring[1024];
+static int fadv_close_buffer_size = 0;
+static size_t fadv_pagesize;
+
+static void fadv_fd_init_func(void)
+{
+	static int fadv_fd_init = 0;
+	if (fadv_fd_init == 0){
+		int i;
+		fadv_fd_init = 1;
+		fadv_pagesize = getpagesize();
+		if (fadv_max_fd == 0){
+			fadv_max_fd = sysconf(_SC_OPEN_MAX) - 20;
+			if (fadv_max_fd < 0)
+				fadv_max_fd = 1;
+			if (fadv_max_fd > 1000)
+				fadv_max_fd = 1000;
+		}
+		for (i=0;i<fadv_max_fd;i++){
+			fadv_fd_pos[i] = 0;
+			fadv_fd_stat[i].st_dev = 0;
+			fadv_fd_stat[i].st_ino = 0;
+			fadv_fd_stat[i].st_size = 0;
+			fadv_core_ptr[i] = NULL;
+		}
+	}
+}
+
+static void fadv_get_core(int fd)
+{
+	struct stat stat;
+	void *pa;
+	size_t pi;
+	fstat(fd,&stat);
+	if ( fadv_fd_stat[fd].st_dev == stat.st_dev && fadv_fd_stat[fd].st_ino == stat.st_ino ) {
+		return;
+	}
+	fadv_fd_stat[fd].st_dev = stat.st_dev;
+	fadv_fd_stat[fd].st_ino = stat.st_ino;
+	fadv_fd_stat[fd].st_size = stat.st_size;
+
+	if (fadv_core_ptr[fd]!=NULL){
+		free (fadv_core_ptr[fd]);
+	}
+
+	pa = mmap((void *)0, stat.st_size, PROT_READ, MAP_SHARED, fd, 0);
+	if (MAP_FAILED == pa) {
+		perror("mmap");
+	} else {
+		fadv_core_ptr[fd] = calloc(1, (stat.st_size+fadv_pagesize)/fadv_pagesize);
+		if ( fadv_core_ptr[fd] == NULL ){
+			perror("calloc");
+		} else {
+			if ( mincore(pa, stat.st_size, (fadv_core_ptr[fd])) != 0){
+				perror("mincore");
+				free(fadv_core_ptr[fd]);
+				fadv_core_ptr[fd]=(unsigned char*)0;
+			} else if (DEBUG_GTE(IO, 4)) {
+//			} else {
+				rprintf(FINFO, "fadv_get_core(fd=%d): ", fd);
+				for (pi = 0; pi <= stat.st_size/fadv_pagesize; pi++) {
+					if ((fadv_core_ptr[fd])[pi]&1) {
+						rprintf(FINFO,"%lu ", (unsigned long)pi);
+					}
+				}
+				rprintf(FINFO,"\n");
+			}
+			munmap(pa, stat.st_size);
+		}
+	}
+}
+
+static void fadv_drop(int fd, int sync)
+{
+	/* trail 1 MB behind in dropping. we do this to make
+	sure that the same block or stripe does not have
+	to be written twice */
+	off_t pos = lseek(fd,0,SEEK_CUR) - 1024*1024;
+	if (fd > fadv_max_fd){
+		return;
+	}
+	if ( fadv_fd_pos[fd] < pos - FADV_BUFFER_SIZE ) {
+		if (sync) {
+			/* if the file is not flushed to disk before calling fadvise,
+			   then the Cache will not be freed and the advise gets ignored
+			   this does give a severe hit on performance. If only there
+			   was a way to mark cache so that it gets release once the data
+			   is written to disk. */
+			fdatasync(fd);
+		}
+		if (fadv_core_ptr[fd] != NULL) {
+			size_t pi;
+			if (pos < fadv_fd_stat[fd].st_size){
+				for (pi = fadv_fd_pos[fd]/fadv_pagesize; pi <= pos/fadv_pagesize; pi++) {
+					if (! (fadv_core_ptr[fd][pi]&1)) {
+						posix_fadvise64(fd, pi*fadv_pagesize, fadv_pagesize, POSIX_FADV_DONTNEED);
+					}
+				}
+			} else {
+				posix_fadvise64(fd, fadv_fd_stat[fd].st_size, pos-fadv_fd_stat[fd].st_size, POSIX_FADV_DONTNEED);
+			}
+		} else {
+			posix_fadvise64(fd, 0, pos, POSIX_FADV_DONTNEED);
+		}
+		fadv_fd_pos[fd] = pos;
+	}
+}
+
+#endif
+
+ssize_t fadv_write(int fd, const void *buf, size_t count)
+{
+	int ret = write(fd, buf, count);
+#ifdef WITH_DROP_CACHE
+	if (drop_cache) {
+		fadv_drop(fd,1);
+	}
+#endif
+	return ret;
+}
+
+ssize_t fadv_read(int fd, void *buf, size_t count)
+{
+	int ret;
+#ifdef WITH_DROP_CACHE
+	if (drop_cache) {
+		fadv_fd_init_func();
+		fadv_get_core(fd);
+	}
+#endif
+	ret = read(fd, buf, count);
+#ifdef WITH_DROP_CACHE
+	if (drop_cache) {
+		fadv_drop(fd,0);
+	}
+#endif
+	return ret;
+}
+
+#ifdef WITH_DROP_CACHE
+void fadv_close_all(void)
+{
+	/* printf ("%i\n",fadv_close_ring_size); */
+	while (fadv_close_ring_size > 0){
+		fdatasync(fadv_close_ring[fadv_close_ring_tail]);
+		if (fadv_core_ptr[fadv_close_ring[fadv_close_ring_tail]]){
+			size_t pi;
+			for (pi = 0; pi <= fadv_fd_stat[fadv_close_ring[fadv_close_ring_tail]].st_size/fadv_pagesize; pi++) {
+				if (!(fadv_core_ptr[fadv_close_ring[fadv_close_ring_tail]][pi]&1)) {
+					posix_fadvise64(fadv_close_ring[fadv_close_ring_tail], pi*fadv_pagesize, fadv_pagesize, POSIX_FADV_DONTNEED);
+				}
+			}
+			/* if the file has grown, drop the rest */
+			//posix_fadvise64(fadv_close_ring[fadv_close_ring_tail], fadv_fd_stat[fadv_close_ring[fadv_close_ring_tail]].st_size,0, POSIX_FADV_DONTNEED);
+
+			free(fadv_core_ptr[fadv_close_ring[fadv_close_ring_tail]]);
+			fadv_core_ptr[fadv_close_ring[fadv_close_ring_tail]] = NULL;
+			fadv_fd_stat[fadv_close_ring[fadv_close_ring_tail]].st_size = 0;
+			fadv_fd_stat[fadv_close_ring[fadv_close_ring_tail]].st_ino = 0;
+			fadv_fd_stat[fadv_close_ring[fadv_close_ring_tail]].st_dev = 0;
+		} else {
+			posix_fadvise64(fadv_close_ring[fadv_close_ring_tail], 0, 0,POSIX_FADV_DONTNEED);
+		}
+		fadv_close_ring_size--;
+		close(fadv_close_ring[fadv_close_ring_tail]);
+		fadv_close_ring_tail = (fadv_close_ring_tail + 1) % fadv_max_fd;
+		fadv_close_buffer_size = 0;
+	}
+}
+
+int fadv_close(int fd)
+{
+	if (drop_cache) {
+		/* if the file is not flushed to disk before calling fadvise,
+		   then the Cache will not be freed and the advise gets ignored
+		   this does give a severe hit on performance. So instead of doing
+		   it right away, we save us a copy of the filehandle and do it
+		   some time before we are out of filehandles. This speeds
+		   up operation for small files massively. It is directly
+		   related to the number of spare file handles you have. */
+		int newfd = dup(fd);
+		off_t pos = lseek(fd,0,SEEK_CUR);
+		fadv_fd_init_func();
+		fadv_core_ptr[newfd] = fadv_core_ptr[fd];
+		fadv_fd_stat[newfd].st_size = fadv_fd_stat[fd].st_size ;
+		fadv_core_ptr[fd] = NULL;
+		fadv_close_buffer_size += pos - fadv_fd_pos[fd];
+		fadv_close_ring[fadv_close_ring_head] = newfd;
+		fadv_close_ring_head = (fadv_close_ring_head + 1) % fadv_max_fd;
+		fadv_close_ring_size ++;
+		if (fadv_close_ring_size == fadv_max_fd || fadv_close_buffer_size > 1024*1024 ){
+			/* it seems fastest to drop things 'in groups' */
+			fadv_close_all();
+		}
+	};
+	return close(fd);
+}
+
+
+#define close(fd) fadv_close(fd)
+#endif
+
 /* Set a fd into nonblocking mode. */
 void set_nonblocking(int fd)
 {
@@ -273,7 +489,7 @@
 
 	total_written = 0;
 	while (len > 0) {
-		int written = write(desc, ptr, len);
+		int written = fadv_write(desc, ptr, len);
 		if (written < 0)  {
 			if (errno == EINTR)
 				continue;
@@ -305,7 +521,7 @@
 		return len;
 
 	do {
-		n_chars = read(desc, ptr, len);
+		n_chars = fadv_read(desc, ptr, len);
 	} while (n_chars < 0 && errno == EINTR);
 
 	return n_chars;