| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* copydir.c |
| 4 |
* copies a directory |
| 5 |
* |
| 6 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 7 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 8 |
* |
| 9 |
* While "xcopy /e /i /q" works fine for copying directories, on Windows XP |
| 10 |
* it requires a Window handle which prevents it from working when invoked |
| 11 |
* as a service. |
| 12 |
* |
| 13 |
* IDENTIFICATION |
| 14 |
* src/backend/storage/file/copydir.c |
| 15 |
* |
| 16 |
*------------------------------------------------------------------------- |
| 17 |
*/ |
| 18 |
|
| 19 |
#include "postgres.h" |
| 20 |
|
| 21 |
#ifdef HAVE_COPYFILE_H |
| 22 |
#include <copyfile.h> |
| 23 |
#endif |
| 24 |
#include <fcntl.h> |
| 25 |
#include <unistd.h> |
| 26 |
|
| 27 |
#include "common/file_utils.h" |
| 28 |
#include "miscadmin.h" |
| 29 |
#include "pgstat.h" |
| 30 |
#include "storage/copydir.h" |
| 31 |
#include "storage/fd.h" |
| 32 |
|
| 33 |
/* GUCs */ |
| 34 |
int file_copy_method = FILE_COPY_METHOD_COPY; |
| 35 |
|
| 36 |
static void clone_file(const char *fromfile, const char *tofile); |
| 37 |
|
| 38 |
/* |
| 39 |
* copydir: copy a directory |
| 40 |
* |
| 41 |
* If recurse is false, subdirectories are ignored. Anything that's not |
| 42 |
* a directory or a regular file is ignored. |
| 43 |
* |
| 44 |
* This function uses the file_copy_method GUC. New uses of this function must |
| 45 |
* be documented in doc/src/sgml/config.sgml. |
| 46 |
*/ |
| 47 |
void |
| 48 |
copydir(const char *fromdir, const char *todir, bool recurse) |
| 49 |
{ |
| 50 |
DIR *xldir; |
| 51 |
struct dirent *xlde; |
| 52 |
char fromfile[MAXPGPATH * 2]; |
| 53 |
char tofile[MAXPGPATH * 2]; |
| 54 |
|
| 55 |
if (MakePGDirectory(todir) != 0) |
| 56 |
ereport(ERROR, |
| 57 |
(errcode_for_file_access(), |
| 58 |
errmsg("could not create directory \"%s\": %m", todir))); |
| 59 |
|
| 60 |
xldir = AllocateDir(fromdir); |
| 61 |
|
| 62 |
while ((xlde = ReadDir(xldir, fromdir)) != NULL) |
| 63 |
{ |
| 64 |
PGFileType xlde_type; |
| 65 |
|
| 66 |
/* If we got a cancel signal during the copy of the directory, quit */ |
| 67 |
CHECK_FOR_INTERRUPTS(); |
| 68 |
|
| 69 |
if (strcmp(xlde->d_name, ".") == 0 || |
| 70 |
strcmp(xlde->d_name, "..") == 0) |
| 71 |
continue; |
| 72 |
|
| 73 |
snprintf(fromfile, sizeof(fromfile), "%s/%s", fromdir, xlde->d_name); |
| 74 |
snprintf(tofile, sizeof(tofile), "%s/%s", todir, xlde->d_name); |
| 75 |
|
| 76 |
xlde_type = get_dirent_type(fromfile, xlde, false, ERROR); |
| 77 |
|
| 78 |
if (xlde_type == PGFILETYPE_DIR) |
| 79 |
{ |
| 80 |
/* recurse to handle subdirectories */ |
| 81 |
if (recurse) |
| 82 |
copydir(fromfile, tofile, true); |
| 83 |
} |
| 84 |
else if (xlde_type == PGFILETYPE_REG) |
| 85 |
{ |
| 86 |
if (file_copy_method == FILE_COPY_METHOD_CLONE) |
| 87 |
clone_file(fromfile, tofile); |
| 88 |
else |
| 89 |
copy_file(fromfile, tofile); |
| 90 |
} |
| 91 |
} |
| 92 |
FreeDir(xldir); |
| 93 |
|
| 94 |
/* |
| 95 |
* Be paranoid here and fsync all files to ensure the copy is really done. |
| 96 |
* But if fsync is disabled, we're done. |
| 97 |
*/ |
| 98 |
if (!enableFsync) |
| 99 |
return; |
| 100 |
|
| 101 |
xldir = AllocateDir(todir); |
| 102 |
|
| 103 |
while ((xlde = ReadDir(xldir, todir)) != NULL) |
| 104 |
{ |
| 105 |
if (strcmp(xlde->d_name, ".") == 0 || |
| 106 |
strcmp(xlde->d_name, "..") == 0) |
| 107 |
continue; |
| 108 |
|
| 109 |
snprintf(tofile, sizeof(tofile), "%s/%s", todir, xlde->d_name); |
| 110 |
|
| 111 |
/* |
| 112 |
* We don't need to sync subdirectories here since the recursive |
| 113 |
* copydir will do it before it returns |
| 114 |
*/ |
| 115 |
if (get_dirent_type(tofile, xlde, false, ERROR) == PGFILETYPE_REG) |
| 116 |
fsync_fname(tofile, false); |
| 117 |
} |
| 118 |
FreeDir(xldir); |
| 119 |
|
| 120 |
/* |
| 121 |
* It's important to fsync the destination directory itself as individual |
| 122 |
* file fsyncs don't guarantee that the directory entry for the file is |
| 123 |
* synced. Recent versions of ext4 have made the window much wider but |
| 124 |
* it's been true for ext3 and other filesystems in the past. |
| 125 |
*/ |
| 126 |
fsync_fname(todir, true); |
| 127 |
} |
| 128 |
|
| 129 |
/* |
| 130 |
* copy one file |
| 131 |
*/ |
| 132 |
void |
| 133 |
copy_file(const char *fromfile, const char *tofile) |
| 134 |
{ |
| 135 |
char *buffer; |
| 136 |
int srcfd; |
| 137 |
int dstfd; |
| 138 |
int nbytes; |
| 139 |
off_t offset; |
| 140 |
off_t flush_offset; |
| 141 |
|
| 142 |
/* Size of copy buffer (read and write requests) */ |
| 143 |
#define COPY_BUF_SIZE (8 * BLCKSZ) |
| 144 |
|
| 145 |
/* |
| 146 |
* Size of data flush requests. It seems beneficial on most platforms to |
| 147 |
* do this every 1MB or so. But macOS, at least with early releases of |
| 148 |
* APFS, is really unfriendly to small mmap/msync requests, so there do it |
| 149 |
* only every 32MB. |
| 150 |
*/ |
| 151 |
#if defined(__darwin__) |
| 152 |
#define FLUSH_DISTANCE (32 * 1024 * 1024) |
| 153 |
#else |
| 154 |
#define FLUSH_DISTANCE (1024 * 1024) |
| 155 |
#endif |
| 156 |
|
| 157 |
/* Use palloc to ensure we get a maxaligned buffer */ |
| 158 |
buffer = palloc(COPY_BUF_SIZE); |
| 159 |
|
| 160 |
/* |
| 161 |
* Open the files |
| 162 |
*/ |
| 163 |
srcfd = OpenTransientFile(fromfile, O_RDONLY | PG_BINARY); |
| 164 |
if (srcfd < 0) |
| 165 |
ereport(ERROR, |
| 166 |
(errcode_for_file_access(), |
| 167 |
errmsg("could not open file \"%s\": %m", fromfile))); |
| 168 |
|
| 169 |
dstfd = OpenTransientFile(tofile, O_RDWR | O_CREAT | O_EXCL | PG_BINARY); |
| 170 |
if (dstfd < 0) |
| 171 |
ereport(ERROR, |
| 172 |
(errcode_for_file_access(), |
| 173 |
errmsg("could not create file \"%s\": %m", tofile))); |
| 174 |
|
| 175 |
/* |
| 176 |
* Do the data copying. |
| 177 |
*/ |
| 178 |
flush_offset = 0; |
| 179 |
for (offset = 0;; offset += nbytes) |
| 180 |
{ |
| 181 |
/* If we got a cancel signal during the copy of the file, quit */ |
| 182 |
CHECK_FOR_INTERRUPTS(); |
| 183 |
|
| 184 |
/* |
| 185 |
* We fsync the files later, but during the copy, flush them every so |
| 186 |
* often to avoid spamming the cache and hopefully get the kernel to |
| 187 |
* start writing them out before the fsync comes. |
| 188 |
*/ |
| 189 |
if (offset - flush_offset >= FLUSH_DISTANCE) |
| 190 |
{ |
| 191 |
pg_flush_data(dstfd, flush_offset, offset - flush_offset); |
| 192 |
flush_offset = offset; |
| 193 |
} |
| 194 |
|
| 195 |
pgstat_report_wait_start(WAIT_EVENT_COPY_FILE_READ); |
| 196 |
nbytes = read(srcfd, buffer, COPY_BUF_SIZE); |
| 197 |
pgstat_report_wait_end(); |
| 198 |
if (nbytes < 0) |
| 199 |
ereport(ERROR, |
| 200 |
(errcode_for_file_access(), |
| 201 |
errmsg("could not read file \"%s\": %m", fromfile))); |
| 202 |
if (nbytes == 0) |
| 203 |
break; |
| 204 |
errno = 0; |
| 205 |
pgstat_report_wait_start(WAIT_EVENT_COPY_FILE_WRITE); |
| 206 |
if ((int) write(dstfd, buffer, nbytes) != nbytes) |
| 207 |
{ |
| 208 |
/* if write didn't set errno, assume problem is no disk space */ |
| 209 |
if (errno == 0) |
| 210 |
errno = ENOSPC; |
| 211 |
ereport(ERROR, |
| 212 |
(errcode_for_file_access(), |
| 213 |
errmsg("could not write to file \"%s\": %m", tofile))); |
| 214 |
} |
| 215 |
pgstat_report_wait_end(); |
| 216 |
} |
| 217 |
|
| 218 |
if (offset > flush_offset) |
| 219 |
pg_flush_data(dstfd, flush_offset, offset - flush_offset); |
| 220 |
|
| 221 |
if (CloseTransientFile(dstfd) != 0) |
| 222 |
ereport(ERROR, |
| 223 |
(errcode_for_file_access(), |
| 224 |
errmsg("could not close file \"%s\": %m", tofile))); |
| 225 |
|
| 226 |
if (CloseTransientFile(srcfd) != 0) |
| 227 |
ereport(ERROR, |
| 228 |
(errcode_for_file_access(), |
| 229 |
errmsg("could not close file \"%s\": %m", fromfile))); |
| 230 |
|
| 231 |
pfree(buffer); |
| 232 |
} |
| 233 |
|
| 234 |
/* |
| 235 |
* clone one file |
| 236 |
*/ |
| 237 |
static void |
| 238 |
clone_file(const char *fromfile, const char *tofile) |
| 239 |
{ |
| 240 |
#if defined(HAVE_COPYFILE) && defined(COPYFILE_CLONE_FORCE) |
| 241 |
if (copyfile(fromfile, tofile, NULL, COPYFILE_CLONE_FORCE) < 0) |
| 242 |
ereport(ERROR, |
| 243 |
(errcode_for_file_access(), |
| 244 |
errmsg("could not clone file \"%s\" to \"%s\": %m", |
| 245 |
fromfile, tofile))); |
| 246 |
#elif defined(HAVE_COPY_FILE_RANGE) |
| 247 |
int srcfd; |
| 248 |
int dstfd; |
| 249 |
ssize_t nbytes; |
| 250 |
|
| 251 |
srcfd = OpenTransientFile(fromfile, O_RDONLY | PG_BINARY); |
| 252 |
if (srcfd < 0) |
| 253 |
ereport(ERROR, |
| 254 |
(errcode_for_file_access(), |
| 255 |
errmsg("could not open file \"%s\": %m", fromfile))); |
| 256 |
|
| 257 |
dstfd = OpenTransientFile(tofile, O_WRONLY | O_CREAT | O_EXCL | PG_BINARY); |
| 258 |
if (dstfd < 0) |
| 259 |
ereport(ERROR, |
| 260 |
(errcode_for_file_access(), |
| 261 |
errmsg("could not create file \"%s\": %m", tofile))); |
| 262 |
|
| 263 |
do |
| 264 |
{ |
| 265 |
/* |
| 266 |
* Don't copy too much at once, so we can check for interrupts from |
| 267 |
* time to time if it falls back to a slow copy. |
| 268 |
*/ |
| 269 |
CHECK_FOR_INTERRUPTS(); |
| 270 |
pgstat_report_wait_start(WAIT_EVENT_COPY_FILE_COPY); |
| 271 |
nbytes = copy_file_range(srcfd, NULL, dstfd, NULL, 1024 * 1024, 0); |
| 272 |
if (nbytes < 0 && errno != EINTR) |
| 273 |
ereport(ERROR, |
| 274 |
(errcode_for_file_access(), |
| 275 |
errmsg("could not clone file \"%s\" to \"%s\": %m", |
| 276 |
fromfile, tofile))); |
| 277 |
pgstat_report_wait_end(); |
| 278 |
} |
| 279 |
while (nbytes != 0); |
| 280 |
|
| 281 |
if (CloseTransientFile(dstfd) != 0) |
| 282 |
ereport(ERROR, |
| 283 |
(errcode_for_file_access(), |
| 284 |
errmsg("could not close file \"%s\": %m", tofile))); |
| 285 |
|
| 286 |
if (CloseTransientFile(srcfd) != 0) |
| 287 |
ereport(ERROR, |
| 288 |
(errcode_for_file_access(), |
| 289 |
errmsg("could not close file \"%s\": %m", fromfile))); |
| 290 |
#else |
| 291 |
/* If there is no CLONE support this function should not be called. */ |
| 292 |
pg_unreachable(); |
| 293 |
#endif |
| 294 |
} |