xref: /openbmc/linux/block/blk-ia-ranges.c (revision 173940b3)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  Block device concurrent positioning ranges.
4  *
5  *  Copyright (C) 2021 Western Digital Corporation or its Affiliates.
6  */
7 #include <linux/kernel.h>
8 #include <linux/blkdev.h>
9 #include <linux/slab.h>
10 #include <linux/init.h>
11 
12 #include "blk.h"
13 
14 static ssize_t
15 blk_ia_range_sector_show(struct blk_independent_access_range *iar,
16 			 char *buf)
17 {
18 	return sprintf(buf, "%llu\n", iar->sector);
19 }
20 
21 static ssize_t
22 blk_ia_range_nr_sectors_show(struct blk_independent_access_range *iar,
23 			     char *buf)
24 {
25 	return sprintf(buf, "%llu\n", iar->nr_sectors);
26 }
27 
28 struct blk_ia_range_sysfs_entry {
29 	struct attribute attr;
30 	ssize_t (*show)(struct blk_independent_access_range *iar, char *buf);
31 };
32 
33 static struct blk_ia_range_sysfs_entry blk_ia_range_sector_entry = {
34 	.attr = { .name = "sector", .mode = 0444 },
35 	.show = blk_ia_range_sector_show,
36 };
37 
38 static struct blk_ia_range_sysfs_entry blk_ia_range_nr_sectors_entry = {
39 	.attr = { .name = "nr_sectors", .mode = 0444 },
40 	.show = blk_ia_range_nr_sectors_show,
41 };
42 
43 static struct attribute *blk_ia_range_attrs[] = {
44 	&blk_ia_range_sector_entry.attr,
45 	&blk_ia_range_nr_sectors_entry.attr,
46 	NULL,
47 };
48 ATTRIBUTE_GROUPS(blk_ia_range);
49 
50 static ssize_t blk_ia_range_sysfs_show(struct kobject *kobj,
51 				      struct attribute *attr, char *buf)
52 {
53 	struct blk_ia_range_sysfs_entry *entry =
54 		container_of(attr, struct blk_ia_range_sysfs_entry, attr);
55 	struct blk_independent_access_range *iar =
56 		container_of(kobj, struct blk_independent_access_range, kobj);
57 
58 	return entry->show(iar, buf);
59 }
60 
61 static const struct sysfs_ops blk_ia_range_sysfs_ops = {
62 	.show	= blk_ia_range_sysfs_show,
63 };
64 
65 /*
66  * Independent access range entries are not freed individually, but alltogether
67  * with struct blk_independent_access_ranges and its array of ranges. Since
68  * kobject_add() takes a reference on the parent kobject contained in
69  * struct blk_independent_access_ranges, the array of independent access range
70  * entries cannot be freed until kobject_del() is called for all entries.
71  * So we do not need to do anything here, but still need this no-op release
72  * operation to avoid complaints from the kobject code.
73  */
74 static void blk_ia_range_sysfs_nop_release(struct kobject *kobj)
75 {
76 }
77 
78 static struct kobj_type blk_ia_range_ktype = {
79 	.sysfs_ops	= &blk_ia_range_sysfs_ops,
80 	.default_groups	= blk_ia_range_groups,
81 	.release	= blk_ia_range_sysfs_nop_release,
82 };
83 
84 /*
85  * This will be executed only after all independent access range entries are
86  * removed with kobject_del(), at which point, it is safe to free everything,
87  * including the array of ranges.
88  */
89 static void blk_ia_ranges_sysfs_release(struct kobject *kobj)
90 {
91 	struct blk_independent_access_ranges *iars =
92 		container_of(kobj, struct blk_independent_access_ranges, kobj);
93 
94 	kfree(iars);
95 }
96 
97 static struct kobj_type blk_ia_ranges_ktype = {
98 	.release	= blk_ia_ranges_sysfs_release,
99 };
100 
101 /**
102  * disk_register_independent_access_ranges - register with sysfs a set of
103  *		independent access ranges
104  * @disk:	Target disk
105  * @new_iars:	New set of independent access ranges
106  *
107  * Register with sysfs a set of independent access ranges for @disk.
108  * If @new_iars is not NULL, this set of ranges is registered and the old set
109  * specified by q->ia_ranges is unregistered. Otherwise, q->ia_ranges is
110  * registered if it is not already.
111  */
112 int disk_register_independent_access_ranges(struct gendisk *disk,
113 				struct blk_independent_access_ranges *new_iars)
114 {
115 	struct request_queue *q = disk->queue;
116 	struct blk_independent_access_ranges *iars;
117 	int i, ret;
118 
119 	lockdep_assert_held(&q->sysfs_dir_lock);
120 	lockdep_assert_held(&q->sysfs_lock);
121 
122 	/* If a new range set is specified, unregister the old one */
123 	if (new_iars) {
124 		if (q->ia_ranges)
125 			disk_unregister_independent_access_ranges(disk);
126 		q->ia_ranges = new_iars;
127 	}
128 
129 	iars = q->ia_ranges;
130 	if (!iars)
131 		return 0;
132 
133 	/*
134 	 * At this point, iars is the new set of sector access ranges that needs
135 	 * to be registered with sysfs.
136 	 */
137 	WARN_ON(iars->sysfs_registered);
138 	ret = kobject_init_and_add(&iars->kobj, &blk_ia_ranges_ktype,
139 				   &q->kobj, "%s", "independent_access_ranges");
140 	if (ret) {
141 		q->ia_ranges = NULL;
142 		kobject_put(&iars->kobj);
143 		return ret;
144 	}
145 
146 	for (i = 0; i < iars->nr_ia_ranges; i++) {
147 		iars->ia_range[i].queue = q;
148 		ret = kobject_init_and_add(&iars->ia_range[i].kobj,
149 					   &blk_ia_range_ktype, &iars->kobj,
150 					   "%d", i);
151 		if (ret) {
152 			while (--i >= 0)
153 				kobject_del(&iars->ia_range[i].kobj);
154 			kobject_del(&iars->kobj);
155 			kobject_put(&iars->kobj);
156 			return ret;
157 		}
158 	}
159 
160 	iars->sysfs_registered = true;
161 
162 	return 0;
163 }
164 
165 void disk_unregister_independent_access_ranges(struct gendisk *disk)
166 {
167 	struct request_queue *q = disk->queue;
168 	struct blk_independent_access_ranges *iars = q->ia_ranges;
169 	int i;
170 
171 	lockdep_assert_held(&q->sysfs_dir_lock);
172 	lockdep_assert_held(&q->sysfs_lock);
173 
174 	if (!iars)
175 		return;
176 
177 	if (iars->sysfs_registered) {
178 		for (i = 0; i < iars->nr_ia_ranges; i++)
179 			kobject_del(&iars->ia_range[i].kobj);
180 		kobject_del(&iars->kobj);
181 		kobject_put(&iars->kobj);
182 	} else {
183 		kfree(iars);
184 	}
185 
186 	q->ia_ranges = NULL;
187 }
188 
189 static struct blk_independent_access_range *
190 disk_find_ia_range(struct blk_independent_access_ranges *iars,
191 		  sector_t sector)
192 {
193 	struct blk_independent_access_range *iar;
194 	int i;
195 
196 	for (i = 0; i < iars->nr_ia_ranges; i++) {
197 		iar = &iars->ia_range[i];
198 		if (sector >= iar->sector &&
199 		    sector < iar->sector + iar->nr_sectors)
200 			return iar;
201 	}
202 
203 	return NULL;
204 }
205 
206 static bool disk_check_ia_ranges(struct gendisk *disk,
207 				struct blk_independent_access_ranges *iars)
208 {
209 	struct blk_independent_access_range *iar, *tmp;
210 	sector_t capacity = get_capacity(disk);
211 	sector_t sector = 0;
212 	int i;
213 
214 	/*
215 	 * While sorting the ranges in increasing LBA order, check that the
216 	 * ranges do not overlap, that there are no sector holes and that all
217 	 * sectors belong to one range.
218 	 */
219 	for (i = 0; i < iars->nr_ia_ranges; i++) {
220 		tmp = disk_find_ia_range(iars, sector);
221 		if (!tmp || tmp->sector != sector) {
222 			pr_warn("Invalid non-contiguous independent access ranges\n");
223 			return false;
224 		}
225 
226 		iar = &iars->ia_range[i];
227 		if (tmp != iar) {
228 			swap(iar->sector, tmp->sector);
229 			swap(iar->nr_sectors, tmp->nr_sectors);
230 		}
231 
232 		sector += iar->nr_sectors;
233 	}
234 
235 	if (sector != capacity) {
236 		pr_warn("Independent access ranges do not match disk capacity\n");
237 		return false;
238 	}
239 
240 	return true;
241 }
242 
243 static bool disk_ia_ranges_changed(struct gendisk *disk,
244 				   struct blk_independent_access_ranges *new)
245 {
246 	struct blk_independent_access_ranges *old = disk->queue->ia_ranges;
247 	int i;
248 
249 	if (!old)
250 		return true;
251 
252 	if (old->nr_ia_ranges != new->nr_ia_ranges)
253 		return true;
254 
255 	for (i = 0; i < old->nr_ia_ranges; i++) {
256 		if (new->ia_range[i].sector != old->ia_range[i].sector ||
257 		    new->ia_range[i].nr_sectors != old->ia_range[i].nr_sectors)
258 			return true;
259 	}
260 
261 	return false;
262 }
263 
264 /**
265  * disk_alloc_independent_access_ranges - Allocate an independent access ranges
266  *                                        data structure
267  * @disk:		target disk
268  * @nr_ia_ranges:	Number of independent access ranges
269  *
270  * Allocate a struct blk_independent_access_ranges structure with @nr_ia_ranges
271  * access range descriptors.
272  */
273 struct blk_independent_access_ranges *
274 disk_alloc_independent_access_ranges(struct gendisk *disk, int nr_ia_ranges)
275 {
276 	struct blk_independent_access_ranges *iars;
277 
278 	iars = kzalloc_node(struct_size(iars, ia_range, nr_ia_ranges),
279 			    GFP_KERNEL, disk->queue->node);
280 	if (iars)
281 		iars->nr_ia_ranges = nr_ia_ranges;
282 	return iars;
283 }
284 EXPORT_SYMBOL_GPL(disk_alloc_independent_access_ranges);
285 
286 /**
287  * disk_set_independent_access_ranges - Set a disk independent access ranges
288  * @disk:	target disk
289  * @iars:	independent access ranges structure
290  *
291  * Set the independent access ranges information of the request queue
292  * of @disk to @iars. If @iars is NULL and the independent access ranges
293  * structure already set is cleared. If there are no differences between
294  * @iars and the independent access ranges structure already set, @iars
295  * is freed.
296  */
297 void disk_set_independent_access_ranges(struct gendisk *disk,
298 				struct blk_independent_access_ranges *iars)
299 {
300 	struct request_queue *q = disk->queue;
301 
302 	if (WARN_ON_ONCE(iars && !iars->nr_ia_ranges)) {
303 		kfree(iars);
304 		iars = NULL;
305 	}
306 
307 	mutex_lock(&q->sysfs_dir_lock);
308 	mutex_lock(&q->sysfs_lock);
309 
310 	if (iars) {
311 		if (!disk_check_ia_ranges(disk, iars)) {
312 			kfree(iars);
313 			iars = NULL;
314 			goto reg;
315 		}
316 
317 		if (!disk_ia_ranges_changed(disk, iars)) {
318 			kfree(iars);
319 			goto unlock;
320 		}
321 	}
322 
323 	/*
324 	 * This may be called for a registered queue. E.g. during a device
325 	 * revalidation. If that is the case, we need to unregister the old
326 	 * set of independent access ranges and register the new set. If the
327 	 * queue is not registered, registration of the device request queue
328 	 * will register the independent access ranges, so only swap in the
329 	 * new set and free the old one.
330 	 */
331 reg:
332 	if (blk_queue_registered(q)) {
333 		disk_register_independent_access_ranges(disk, iars);
334 	} else {
335 		swap(q->ia_ranges, iars);
336 		kfree(iars);
337 	}
338 
339 unlock:
340 	mutex_unlock(&q->sysfs_lock);
341 	mutex_unlock(&q->sysfs_dir_lock);
342 }
343 EXPORT_SYMBOL_GPL(disk_set_independent_access_ranges);
344