xref: /qemu/hw/block/block.c (revision 7653b1ea)
1 /*
2  * Common code for block device models
3  *
4  * Copyright (C) 2012 Red Hat, Inc.
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  */
9 
10 #include "qemu/osdep.h"
11 #include "block/block_int-common.h"
12 #include "sysemu/blockdev.h"
13 #include "sysemu/block-backend.h"
14 #include "hw/block/block.h"
15 #include "qapi/error.h"
16 #include "qapi/qapi-types-block.h"
17 
18 /*
19  * Read the non-zeroes parts of @blk into @buf
20  * Reading all of the @blk is expensive if the zeroes parts of @blk
21  * is large enough. Therefore check the block status and only write
22  * the non-zeroes block into @buf.
23  *
24  * Return 0 on success, non-zero on error.
25  */
26 static int blk_pread_nonzeroes(BlockBackend *blk, hwaddr size, void *buf)
27 {
28     int ret;
29     int64_t bytes, offset = 0;
30     BlockDriverState *bs = blk_bs(blk);
31 
32     for (;;) {
33         bytes = MIN(size - offset, BDRV_REQUEST_MAX_BYTES);
34         if (bytes <= 0) {
35             return 0;
36         }
37         ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL);
38         if (ret < 0) {
39             return ret;
40         }
41         if (!(ret & BDRV_BLOCK_ZERO)) {
42             ret = blk_pread(blk, offset, bytes, (uint8_t *) buf + offset, 0);
43             if (ret < 0) {
44                 return ret;
45             }
46         }
47         offset += bytes;
48     }
49 }
50 
51 /*
52  * Read the entire contents of @blk into @buf.
53  * @blk's contents must be @size bytes, and @size must be at most
54  * BDRV_REQUEST_MAX_BYTES.
55  * On success, return true.
56  * On failure, store an error through @errp and return false.
57  *
58  * This function not intended for actual block devices, which read on
59  * demand.  It's for things like memory devices that (ab)use a block
60  * backend to provide persistence.
61  */
62 bool blk_check_size_and_read_all(BlockBackend *blk, DeviceState *dev,
63                                  void *buf, hwaddr size, Error **errp)
64 {
65     int64_t blk_len;
66     int ret;
67     g_autofree char *dev_id = NULL;
68 
69     blk_len = blk_getlength(blk);
70     if (blk_len < 0) {
71         error_setg_errno(errp, -blk_len,
72                          "can't get size of %s block backend", blk_name(blk));
73         return false;
74     }
75     if (blk_len != size) {
76         dev_id = qdev_get_human_name(dev);
77         error_setg(errp, "%s device '%s' requires %" HWADDR_PRIu
78                    " bytes, %s block backend provides %" PRIu64 " bytes",
79                    object_get_typename(OBJECT(dev)), dev_id, size,
80                    blk_name(blk), blk_len);
81         return false;
82     }
83 
84     /*
85      * We could loop for @size > BDRV_REQUEST_MAX_BYTES, but if we
86      * ever get to the point we want to read *gigabytes* here, we
87      * should probably rework the device to be more like an actual
88      * block device and read only on demand.
89      */
90     assert(size <= BDRV_REQUEST_MAX_BYTES);
91     ret = blk_pread_nonzeroes(blk, size, buf);
92     if (ret < 0) {
93         dev_id = qdev_get_human_name(dev);
94         error_setg_errno(errp, -ret, "can't read %s block backend"
95                          " for %s device '%s'",
96                          blk_name(blk), object_get_typename(OBJECT(dev)),
97                          dev_id);
98         return false;
99     }
100     return true;
101 }
102 
103 bool blkconf_blocksizes(BlockConf *conf, Error **errp)
104 {
105     BlockBackend *blk = conf->blk;
106     BlockSizes blocksizes;
107     BlockDriverState *bs;
108     bool use_blocksizes;
109     bool use_bs;
110 
111     switch (conf->backend_defaults) {
112     case ON_OFF_AUTO_AUTO:
113         use_blocksizes = !blk_probe_blocksizes(blk, &blocksizes);
114         use_bs = false;
115         break;
116 
117     case ON_OFF_AUTO_ON:
118         use_blocksizes = !blk_probe_blocksizes(blk, &blocksizes);
119         bs = blk_bs(blk);
120         use_bs = bs;
121         break;
122 
123     case ON_OFF_AUTO_OFF:
124         use_blocksizes = false;
125         use_bs = false;
126         break;
127 
128     default:
129         abort();
130     }
131 
132     /* fill in detected values if they are not defined via qemu command line */
133     if (!conf->physical_block_size) {
134         if (use_blocksizes) {
135            conf->physical_block_size = blocksizes.phys;
136         } else {
137             conf->physical_block_size = BDRV_SECTOR_SIZE;
138         }
139     }
140     if (!conf->logical_block_size) {
141         if (use_blocksizes) {
142             conf->logical_block_size = blocksizes.log;
143         } else {
144             conf->logical_block_size = BDRV_SECTOR_SIZE;
145         }
146     }
147     if (use_bs) {
148         if (!conf->opt_io_size) {
149             conf->opt_io_size = bs->bl.opt_transfer;
150         }
151         if (conf->discard_granularity == -1) {
152             if (bs->bl.pdiscard_alignment) {
153                 conf->discard_granularity = bs->bl.pdiscard_alignment;
154             } else if (bs->bl.request_alignment != 1) {
155                 conf->discard_granularity = bs->bl.request_alignment;
156             }
157         }
158     }
159 
160     if (conf->logical_block_size > conf->physical_block_size) {
161         error_setg(errp,
162                    "logical_block_size > physical_block_size not supported");
163         return false;
164     }
165 
166     if (!QEMU_IS_ALIGNED(conf->min_io_size, conf->logical_block_size)) {
167         error_setg(errp,
168                    "min_io_size must be a multiple of logical_block_size");
169         return false;
170     }
171 
172     /*
173      * all devices which support min_io_size (scsi and virtio-blk) expose it to
174      * the guest as a uint16_t in units of logical blocks
175      */
176     if (conf->min_io_size / conf->logical_block_size > UINT16_MAX) {
177         error_setg(errp, "min_io_size must not exceed %u logical blocks",
178                    UINT16_MAX);
179         return false;
180     }
181 
182     if (!QEMU_IS_ALIGNED(conf->opt_io_size, conf->logical_block_size)) {
183         error_setg(errp,
184                    "opt_io_size must be a multiple of logical_block_size");
185         return false;
186     }
187 
188     if (conf->discard_granularity != -1 &&
189         !QEMU_IS_ALIGNED(conf->discard_granularity,
190                          conf->logical_block_size)) {
191         error_setg(errp, "discard_granularity must be "
192                    "a multiple of logical_block_size");
193         return false;
194     }
195 
196     return true;
197 }
198 
199 bool blkconf_apply_backend_options(BlockConf *conf, bool readonly,
200                                    bool resizable, Error **errp)
201 {
202     BlockBackend *blk = conf->blk;
203     BlockdevOnError rerror, werror;
204     uint64_t perm, shared_perm;
205     bool wce;
206     int ret;
207 
208     perm = BLK_PERM_CONSISTENT_READ;
209     if (!readonly) {
210         perm |= BLK_PERM_WRITE;
211     }
212 
213     shared_perm = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
214     if (resizable) {
215         shared_perm |= BLK_PERM_RESIZE;
216     }
217     if (conf->share_rw) {
218         shared_perm |= BLK_PERM_WRITE;
219     }
220 
221     ret = blk_set_perm(blk, perm, shared_perm, errp);
222     if (ret < 0) {
223         return false;
224     }
225 
226     switch (conf->wce) {
227     case ON_OFF_AUTO_ON:    wce = true; break;
228     case ON_OFF_AUTO_OFF:   wce = false; break;
229     case ON_OFF_AUTO_AUTO:  wce = blk_enable_write_cache(blk); break;
230     default:
231         abort();
232     }
233 
234     rerror = conf->rerror;
235     if (rerror == BLOCKDEV_ON_ERROR_AUTO) {
236         rerror = blk_get_on_error(blk, true);
237     }
238 
239     werror = conf->werror;
240     if (werror == BLOCKDEV_ON_ERROR_AUTO) {
241         werror = blk_get_on_error(blk, false);
242     }
243 
244     blk_set_enable_write_cache(blk, wce);
245     blk_set_on_error(blk, rerror, werror);
246 
247     block_acct_setup(blk_get_stats(blk), conf->account_invalid,
248                      conf->account_failed);
249     return true;
250 }
251 
252 bool blkconf_geometry(BlockConf *conf, int *ptrans,
253                       unsigned cyls_max, unsigned heads_max, unsigned secs_max,
254                       Error **errp)
255 {
256     if (!conf->cyls && !conf->heads && !conf->secs) {
257         hd_geometry_guess(conf->blk,
258                           &conf->cyls, &conf->heads, &conf->secs,
259                           ptrans);
260     } else if (ptrans && *ptrans == BIOS_ATA_TRANSLATION_AUTO) {
261         *ptrans = hd_bios_chs_auto_trans(conf->cyls, conf->heads, conf->secs);
262     }
263     if (conf->cyls || conf->heads || conf->secs) {
264         if (conf->cyls < 1 || conf->cyls > cyls_max) {
265             error_setg(errp, "cyls must be between 1 and %u", cyls_max);
266             return false;
267         }
268         if (conf->heads < 1 || conf->heads > heads_max) {
269             error_setg(errp, "heads must be between 1 and %u", heads_max);
270             return false;
271         }
272         if (conf->secs < 1 || conf->secs > secs_max) {
273             error_setg(errp, "secs must be between 1 and %u", secs_max);
274             return false;
275         }
276     }
277     return true;
278 }
279