xref: /kvmtool/disk/core.c (revision fea74936f1c8621ebed26a4b3f40b9e045a4bc81)
1 #include "kvm/disk-image.h"
2 #include "kvm/qcow.h"
3 #include "kvm/virtio-blk.h"
4 
5 #include <linux/err.h>
6 #include <sys/eventfd.h>
7 #include <sys/poll.h>
8 
9 #define AIO_MAX 256
10 
11 int debug_iodelay;
12 
13 #ifdef CONFIG_HAS_AIO
14 static void *disk_image__thread(void *param)
15 {
16 	struct disk_image *disk = param;
17 	struct io_event event[AIO_MAX];
18 	struct timespec notime = {0};
19 	int nr, i;
20 	u64 dummy;
21 
22 	while (read(disk->evt, &dummy, sizeof(dummy)) > 0) {
23 		nr = io_getevents(disk->ctx, 1, ARRAY_SIZE(event), event, &notime);
24 		for (i = 0; i < nr; i++)
25 			disk->disk_req_cb(event[i].data, event[i].res);
26 	}
27 
28 	return NULL;
29 }
30 #endif
31 
32 struct disk_image *disk_image__new(int fd, u64 size,
33 				   struct disk_image_operations *ops,
34 				   int use_mmap)
35 {
36 	struct disk_image *disk;
37 	int r;
38 
39 	disk = malloc(sizeof *disk);
40 	if (!disk)
41 		return ERR_PTR(-ENOMEM);
42 
43 	*disk = (struct disk_image) {
44 		.fd	= fd,
45 		.size	= size,
46 		.ops	= ops,
47 	};
48 
49 	if (use_mmap == DISK_IMAGE_MMAP) {
50 		/*
51 		 * The write to disk image will be discarded
52 		 */
53 		disk->priv = mmap(NULL, size, PROT_RW, MAP_PRIVATE | MAP_NORESERVE, fd, 0);
54 		if (disk->priv == MAP_FAILED) {
55 			r = -errno;
56 			free(disk);
57 			return ERR_PTR(r);
58 		}
59 	}
60 
61 #ifdef CONFIG_HAS_AIO
62 	if (disk) {
63 		pthread_t thread;
64 
65 		disk->evt = eventfd(0, 0);
66 		io_setup(AIO_MAX, &disk->ctx);
67 		r = pthread_create(&thread, NULL, disk_image__thread, disk);
68 		if (r) {
69 			r = -errno;
70 			free(disk);
71 			return ERR_PTR(r);
72 		}
73 	}
74 #endif
75 	return disk;
76 }
77 
78 struct disk_image *disk_image__open(const char *filename, bool readonly)
79 {
80 	struct disk_image *disk;
81 	struct stat st;
82 	int fd;
83 
84 	if (stat(filename, &st) < 0)
85 		return ERR_PTR(-errno);
86 
87 	/* blk device ?*/
88 	disk = blkdev__probe(filename, &st);
89 	if (!IS_ERR_OR_NULL(disk))
90 		return disk;
91 
92 	fd = open(filename, readonly ? O_RDONLY : O_RDWR);
93 	if (fd < 0)
94 		return ERR_PTR(fd);
95 
96 	/* qcow image ?*/
97 	disk = qcow_probe(fd, true);
98 	if (!IS_ERR_OR_NULL(disk)) {
99 		pr_warning("Forcing read-only support for QCOW");
100 		return disk;
101 	}
102 
103 	/* raw image ?*/
104 	disk = raw_image__probe(fd, &st, readonly);
105 	if (!IS_ERR_OR_NULL(disk))
106 		return disk;
107 
108 	if (close(fd) < 0)
109 		pr_warning("close() failed");
110 
111 	return ERR_PTR(-ENOSYS);
112 }
113 
114 struct disk_image **disk_image__open_all(const char **filenames,
115 					 bool *readonly, int count)
116 {
117 	struct disk_image **disks;
118 	int i;
119 	void *err;
120 
121 	if (!count)
122 		return ERR_PTR(-EINVAL);
123 	if (count > MAX_DISK_IMAGES)
124 		return ERR_PTR(-ENOSPC);
125 
126 	disks = calloc(count, sizeof(*disks));
127 	if (!disks)
128 		return ERR_PTR(-ENOMEM);
129 
130 	for (i = 0; i < count; i++) {
131 		if (!filenames[i])
132 			continue;
133 
134 		disks[i] = disk_image__open(filenames[i], readonly[i]);
135 		if (IS_ERR_OR_NULL(disks[i])) {
136 			pr_err("Loading disk image '%s' failed", filenames[i]);
137 			err = disks[i];
138 			goto error;
139 		}
140 	}
141 
142 	return disks;
143 error:
144 	for (i = 0; i < count; i++)
145 		if (!IS_ERR_OR_NULL(disks[i]))
146 			disk_image__close(disks[i]);
147 
148 	free(disks);
149 	return err;
150 }
151 
152 int disk_image__flush(struct disk_image *disk)
153 {
154 	if (disk->ops->flush)
155 		return disk->ops->flush(disk);
156 
157 	return fsync(disk->fd);
158 }
159 
160 int disk_image__close(struct disk_image *disk)
161 {
162 	/* If there was no disk image then there's nothing to do: */
163 	if (!disk)
164 		return 0;
165 
166 	if (disk->ops->close)
167 		return disk->ops->close(disk);
168 
169 	if (close(disk->fd) < 0)
170 		pr_warning("close() failed");
171 
172 	free(disk);
173 
174 	return 0;
175 }
176 
177 int disk_image__close_all(struct disk_image **disks, int count)
178 {
179 	while (count)
180 		disk_image__close(disks[--count]);
181 
182 	free(disks);
183 
184 	return 0;
185 }
186 
187 /*
188  * Fill iov with disk data, starting from sector 'sector'.
189  * Return amount of bytes read.
190  */
191 ssize_t disk_image__read(struct disk_image *disk, u64 sector,
192 			 const struct iovec *iov, int iovcount, void *param)
193 {
194 	ssize_t total = 0;
195 
196 	if (debug_iodelay)
197 		msleep(debug_iodelay);
198 
199 	if (disk->ops->read) {
200 		total = disk->ops->read(disk, sector, iov, iovcount, param);
201 		if (total < 0) {
202 			pr_info("disk_image__read error: total=%ld\n", (long)total);
203 			return total;
204 		}
205 	}
206 
207 	if (!disk->async && disk->disk_req_cb)
208 		disk->disk_req_cb(param, total);
209 
210 	return total;
211 }
212 
213 /*
214  * Write iov to disk, starting from sector 'sector'.
215  * Return amount of bytes written.
216  */
217 ssize_t disk_image__write(struct disk_image *disk, u64 sector,
218 			  const struct iovec *iov, int iovcount, void *param)
219 {
220 	ssize_t total = 0;
221 
222 	if (debug_iodelay)
223 		msleep(debug_iodelay);
224 
225 	if (disk->ops->write) {
226 		/*
227 		 * Try writev based operation first
228 		 */
229 
230 		total = disk->ops->write(disk, sector, iov, iovcount, param);
231 		if (total < 0) {
232 			pr_info("disk_image__write error: total=%ld\n", (long)total);
233 			return total;
234 		}
235 	} else {
236 		/* Do nothing */
237 	}
238 
239 	if (!disk->async && disk->disk_req_cb)
240 		disk->disk_req_cb(param, total);
241 
242 	return total;
243 }
244 
245 ssize_t disk_image__get_serial(struct disk_image *disk, void *buffer, ssize_t *len)
246 {
247 	struct stat st;
248 	int r;
249 
250 	r = fstat(disk->fd, &st);
251 	if (r)
252 		return r;
253 
254 	*len = snprintf(buffer, *len, "%llu%llu%llu",
255 			(u64)st.st_dev, (u64)st.st_rdev, (u64)st.st_ino);
256 	return *len;
257 }
258 
259 void disk_image__set_callback(struct disk_image *disk,
260 			      void (*disk_req_cb)(void *param, long len))
261 {
262 	disk->disk_req_cb = disk_req_cb;
263 }
264