4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * All rights reserved. This program and the accompanying materials
7 * are made available under the terms of the GNU Lesser General Public License
8 * (LGPL) version 2.1 or (at your discretion) any later version.
9 * (LGPL) version 2.1 accompanies this distribution, and is available at
10 * http://www.gnu.org/licenses/lgpl-2.1.html
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
20 * This file is part of Lustre, http://www.lustre.org/
22 * lustre/utils/liblustreapi_mirror.c
24 * Copyright (c) 2017, Intel Corporation.
26 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
33 #include <sys/ioctl.h>
40 #include <sys/types.h>
41 #include <sys/xattr.h>
43 #include <sys/param.h>
45 #include <libcfs/util/ioctl.h>
46 #include <lustre/lustreapi.h>
47 #include <linux/lustre/lustre_ioctl.h>
50 * Set the mirror id for the opening file pointed by @fd, once the mirror
51 * is set successfully, the policy to choose mirrors will be disabed and the
52 * following I/O from this file descriptor will be led to this dedicated
54 * If @id is zero, it will clear the mirror id setting.
56 * \param fd file descriptor, must be opened with O_DIRECT
59 * \retval 0 on success.
60 * \retval -errno on failure.
62 int llapi_mirror_set(int fd, unsigned int id)
67 rc = ioctl(fd, LL_IOC_FLR_SET_MIRROR, id);
76 /* in the current implementation, llite doesn't verify if the mirror
77 * id is valid, it has to be verified in an I/O context so the fstat()
78 * call is to verify that the mirror id is correct. */
79 rc = fstat(fd, &stbuf);
83 (void) ioctl(fd, LL_IOC_FLR_SET_MIRROR, 0);
90 * Clear mirror id setting.
92 * \See llapi_mirror_set() for details.
94 int llapi_mirror_clear(int fd)
96 return llapi_mirror_set(fd, 0);
100 * Read data from a specified mirror with @id. This function won't read
101 * partial read result; either file end is reached, or number of @count bytes
102 * is read, or an error will be returned.
104 * \param fd file descriptor, should be opened with O_DIRECT
105 * \param id mirror id to be read from
106 * \param buf read buffer
107 * \param count number of bytes to be read
108 * \param pos file postion where the read starts
110 * \result >= 0 Number of bytes has been read
111 * \result < 0 The last seen error
113 ssize_t llapi_mirror_read(int fd, unsigned int id, void *buf, size_t count,
116 size_t page_size = sysconf(_SC_PAGESIZE);
120 rc = llapi_mirror_set(fd, id);
127 bytes_read = pread(fd, buf, count, pos);
128 if (!bytes_read) /* end of file */
131 if (bytes_read < 0) {
136 result += bytes_read;
141 if (bytes_read & (page_size - 1)) /* end of file */
145 (void) llapi_mirror_clear(fd);
150 ssize_t llapi_mirror_write(int fd, unsigned int id, const void *buf,
151 size_t count, off_t pos)
153 size_t page_size = sysconf(_SC_PAGESIZE);
157 if (((unsigned long)buf & (page_size - 1)) || pos & (page_size - 1))
160 rc = llapi_mirror_set(fd, id);
165 ssize_t bytes_written;
167 if (pos & (page_size - 1)) {
172 bytes_written = pwrite(fd, buf, count, pos);
173 if (bytes_written < 0) {
178 result += bytes_written;
179 pos += bytes_written;
180 buf += bytes_written;
181 count -= bytes_written;
184 (void) llapi_mirror_clear(fd);
189 int llapi_mirror_truncate(int fd, unsigned int id, off_t length)
193 rc = llapi_mirror_set(fd, id);
197 rc = ftruncate(fd, length);
201 (void) llapi_mirror_clear(fd);
206 int llapi_mirror_punch(int fd, unsigned int id, off_t start, size_t length)
210 rc = llapi_mirror_set(fd, id);
214 rc = llapi_hole_punch(fd, start, length);
215 (void) llapi_mirror_clear(fd);
220 bool llapi_mirror_is_sparse(int fd, unsigned int id)
225 rc = llapi_mirror_set(fd, id);
229 sparse = llapi_file_is_sparse(fd);
230 (void) llapi_mirror_clear(fd);
236 * Seek data in a specified mirror with @id. This function looks for the
237 * first data segment from given offset and returns its offset and length
239 * \param fd file descriptor, should be opened with O_DIRECT
240 * \param id mirror id to be read from
241 * \param pos position for start data seek from
242 * \param size size of data segment found
244 * \result >= 0 Number of bytes has been read
245 * \result < 0 The last seen error
247 off_t llapi_mirror_data_seek(int fd, unsigned int id, off_t pos, size_t *size)
252 rc = llapi_mirror_set(fd, id);
256 data_off = llapi_data_seek(fd, pos, size);
257 (void) llapi_mirror_clear(fd);
263 * Copy data contents from source mirror @src to multiple destinations
264 * pointed by @dst. The destination array @dst will be altered to store
265 * successfully copied mirrors.
267 * \param fd file descriptor, should be opened with O_DIRECT
268 * \param src source mirror id, usually a valid mirror
269 * \param dst an array of destination mirror ids
270 * \param count number of elements in array @dst
272 * \result > 0 Number of mirrors successfully copied
273 * \result < 0 The last seen error
275 ssize_t llapi_mirror_copy_many(int fd, __u16 src, __u16 *dst, size_t count)
277 const size_t buflen = 4 * 1024 * 1024; /* 4M */
281 size_t page_size = sysconf(_SC_PAGESIZE);
292 rc = posix_memalign(&buf, page_size, buflen);
293 if (rc) /* error code is returned directly */
296 sparse = llapi_mirror_is_sparse(fd, src);
300 /* for sparse src we have to be sure that dst has no
301 * data in src holes, so truncate it first
303 for (i = 0; i < nr; i++) {
304 rc = llapi_mirror_truncate(fd, dst[i], pos);
307 /* exclude the failed one */
320 size_t to_write, to_read;
322 if (sparse && pos >= data_end) {
325 data_off = llapi_mirror_data_seek(fd, src, pos,
328 /* Non-fatal, switch to full copy */
333 /* hole at the end of file, set pos to the
334 * data_off, so truncate block at the end
335 * will set final dst size.
341 data_end = data_off + data_size;
343 pos = data_off & ~(page_size - 1);
344 data_end = ((data_end - 1) | (page_size - 1)) + 1;
345 to_read = MIN(data_end - pos, buflen);
350 bytes_read = llapi_mirror_read(fd, src, buf, to_read, pos);
351 if (!bytes_read) { /* end of file */
353 } else if (bytes_read < 0) {
359 /* round up to page align to make direct IO happy.
360 * this implies the last segment to write. */
361 to_write = ((bytes_read - 1) | (page_size - 1)) + 1;
363 for (i = 0; i < nr; i++) {
366 written = llapi_mirror_write(fd, dst[i], buf,
371 /* this mirror is not written succesfully,
372 * get rid of it from the array */
377 assert(written == to_write);
380 eof = bytes_read < to_read;
386 for (i = 0; i < nr; i++) {
387 rc = llapi_mirror_truncate(fd, dst[i], pos);
391 /* exclude the failed one */
399 return nr > 0 ? nr : result;
403 * Copy data contents from source mirror @src to target mirror @dst.
405 * \param fd file descriptor, should be opened with O_DIRECT
406 * \param src source mirror id, usually a valid mirror
407 * \param dst mirror id of copy destination
408 * \param pos start file pos
409 * \param count number of bytes to be copied
411 * \result > 0 Number of mirrors successfully copied
412 * \result < 0 The last seen error
414 int llapi_mirror_copy(int fd, unsigned int src, unsigned int dst, off_t pos,
417 const size_t buflen = 4 * 1024 * 1024; /* 4M */
419 size_t page_size = sysconf(_SC_PAGESIZE);
426 if (pos & (page_size - 1) || !dst)
429 if (count != OBD_OBJECT_EOF && count & (page_size - 1))
432 rc = posix_memalign(&buf, page_size, buflen);
433 if (rc) /* error code is returned directly */
436 while (result < count) {
437 ssize_t bytes_read, bytes_written;
438 size_t to_read, to_write;
440 to_read = MIN(buflen, count - result);
442 bytes_read = pread(fd, buf, to_read, pos);
444 bytes_read = llapi_mirror_read(fd, src, buf, to_read,
446 if (!bytes_read) { /* end of file */
448 } else if (bytes_read < 0) {
453 /* round up to page align to make direct IO happy.
454 * this implies the last segment to write. */
455 to_write = (bytes_read + page_size - 1) & ~(page_size - 1);
457 bytes_written = llapi_mirror_write(fd, dst, buf, to_write,
459 if (bytes_written < 0) {
460 result = bytes_written;
464 assert(bytes_written == to_write);
467 result += bytes_read;
469 if (bytes_read < to_read) /* short read occurred */
475 if (result > 0 && pos & (page_size - 1)) {
476 rc = llapi_mirror_truncate(fd, dst, pos);