blob: 3405b4ee1757e3deb196c93e49b9641dad8d56fb
1 | #include <linux/swap_cgroup.h> |
2 | #include <linux/vmalloc.h> |
3 | #include <linux/mm.h> |
4 | |
5 | #include <linux/swapops.h> /* depends on mm.h include */ |
6 | |
7 | static DEFINE_MUTEX(swap_cgroup_mutex); |
8 | struct swap_cgroup_ctrl { |
9 | struct page **map; |
10 | unsigned long length; |
11 | spinlock_t lock; |
12 | }; |
13 | |
14 | static struct swap_cgroup_ctrl swap_cgroup_ctrl[MAX_SWAPFILES]; |
15 | |
16 | struct swap_cgroup { |
17 | unsigned short id; |
18 | }; |
19 | #define SC_PER_PAGE (PAGE_SIZE/sizeof(struct swap_cgroup)) |
20 | |
21 | /* |
22 | * SwapCgroup implements "lookup" and "exchange" operations. |
23 | * In typical usage, this swap_cgroup is accessed via memcg's charge/uncharge |
24 | * against SwapCache. At swap_free(), this is accessed directly from swap. |
25 | * |
26 | * This means, |
27 | * - we have no race in "exchange" when we're accessed via SwapCache because |
28 | * SwapCache(and its swp_entry) is under lock. |
29 | * - When called via swap_free(), there is no user of this entry and no race. |
30 | * Then, we don't need lock around "exchange". |
31 | * |
32 | * TODO: we can push these buffers out to HIGHMEM. |
33 | */ |
34 | |
35 | /* |
36 | * allocate buffer for swap_cgroup. |
37 | */ |
38 | static int swap_cgroup_prepare(int type) |
39 | { |
40 | struct page *page; |
41 | struct swap_cgroup_ctrl *ctrl; |
42 | unsigned long idx, max; |
43 | |
44 | ctrl = &swap_cgroup_ctrl[type]; |
45 | |
46 | for (idx = 0; idx < ctrl->length; idx++) { |
47 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
48 | if (!page) |
49 | goto not_enough_page; |
50 | ctrl->map[idx] = page; |
51 | |
52 | if (!(idx % SWAP_CLUSTER_MAX)) |
53 | cond_resched(); |
54 | } |
55 | return 0; |
56 | not_enough_page: |
57 | max = idx; |
58 | for (idx = 0; idx < max; idx++) |
59 | __free_page(ctrl->map[idx]); |
60 | |
61 | return -ENOMEM; |
62 | } |
63 | |
64 | static struct swap_cgroup *lookup_swap_cgroup(swp_entry_t ent, |
65 | struct swap_cgroup_ctrl **ctrlp) |
66 | { |
67 | pgoff_t offset = swp_offset(ent); |
68 | struct swap_cgroup_ctrl *ctrl; |
69 | struct page *mappage; |
70 | struct swap_cgroup *sc; |
71 | |
72 | ctrl = &swap_cgroup_ctrl[swp_type(ent)]; |
73 | if (ctrlp) |
74 | *ctrlp = ctrl; |
75 | |
76 | mappage = ctrl->map[offset / SC_PER_PAGE]; |
77 | sc = page_address(mappage); |
78 | return sc + offset % SC_PER_PAGE; |
79 | } |
80 | |
81 | /** |
82 | * swap_cgroup_cmpxchg - cmpxchg mem_cgroup's id for this swp_entry. |
83 | * @ent: swap entry to be cmpxchged |
84 | * @old: old id |
85 | * @new: new id |
86 | * |
87 | * Returns old id at success, 0 at failure. |
88 | * (There is no mem_cgroup using 0 as its id) |
89 | */ |
90 | unsigned short swap_cgroup_cmpxchg(swp_entry_t ent, |
91 | unsigned short old, unsigned short new) |
92 | { |
93 | struct swap_cgroup_ctrl *ctrl; |
94 | struct swap_cgroup *sc; |
95 | unsigned long flags; |
96 | unsigned short retval; |
97 | |
98 | sc = lookup_swap_cgroup(ent, &ctrl); |
99 | |
100 | spin_lock_irqsave(&ctrl->lock, flags); |
101 | retval = sc->id; |
102 | if (retval == old) |
103 | sc->id = new; |
104 | else |
105 | retval = 0; |
106 | spin_unlock_irqrestore(&ctrl->lock, flags); |
107 | return retval; |
108 | } |
109 | |
110 | /** |
111 | * swap_cgroup_record - record mem_cgroup for this swp_entry. |
112 | * @ent: swap entry to be recorded into |
113 | * @id: mem_cgroup to be recorded |
114 | * |
115 | * Returns old value at success, 0 at failure. |
116 | * (Of course, old value can be 0.) |
117 | */ |
118 | unsigned short swap_cgroup_record(swp_entry_t ent, unsigned short id) |
119 | { |
120 | struct swap_cgroup_ctrl *ctrl; |
121 | struct swap_cgroup *sc; |
122 | unsigned short old; |
123 | unsigned long flags; |
124 | |
125 | sc = lookup_swap_cgroup(ent, &ctrl); |
126 | |
127 | spin_lock_irqsave(&ctrl->lock, flags); |
128 | old = sc->id; |
129 | sc->id = id; |
130 | spin_unlock_irqrestore(&ctrl->lock, flags); |
131 | |
132 | return old; |
133 | } |
134 | |
135 | /** |
136 | * lookup_swap_cgroup_id - lookup mem_cgroup id tied to swap entry |
137 | * @ent: swap entry to be looked up. |
138 | * |
139 | * Returns ID of mem_cgroup at success. 0 at failure. (0 is invalid ID) |
140 | */ |
141 | unsigned short lookup_swap_cgroup_id(swp_entry_t ent) |
142 | { |
143 | return lookup_swap_cgroup(ent, NULL)->id; |
144 | } |
145 | |
146 | int swap_cgroup_swapon(int type, unsigned long max_pages) |
147 | { |
148 | void *array; |
149 | unsigned long array_size; |
150 | unsigned long length; |
151 | struct swap_cgroup_ctrl *ctrl; |
152 | |
153 | if (!do_swap_account) |
154 | return 0; |
155 | |
156 | length = DIV_ROUND_UP(max_pages, SC_PER_PAGE); |
157 | array_size = length * sizeof(void *); |
158 | |
159 | array = vzalloc(array_size); |
160 | if (!array) |
161 | goto nomem; |
162 | |
163 | ctrl = &swap_cgroup_ctrl[type]; |
164 | mutex_lock(&swap_cgroup_mutex); |
165 | ctrl->length = length; |
166 | ctrl->map = array; |
167 | spin_lock_init(&ctrl->lock); |
168 | if (swap_cgroup_prepare(type)) { |
169 | /* memory shortage */ |
170 | ctrl->map = NULL; |
171 | ctrl->length = 0; |
172 | mutex_unlock(&swap_cgroup_mutex); |
173 | vfree(array); |
174 | goto nomem; |
175 | } |
176 | mutex_unlock(&swap_cgroup_mutex); |
177 | |
178 | return 0; |
179 | nomem: |
180 | pr_info("couldn't allocate enough memory for swap_cgroup\n"); |
181 | pr_info("swap_cgroup can be disabled by swapaccount=0 boot option\n"); |
182 | return -ENOMEM; |
183 | } |
184 | |
185 | void swap_cgroup_swapoff(int type) |
186 | { |
187 | struct page **map; |
188 | unsigned long i, length; |
189 | struct swap_cgroup_ctrl *ctrl; |
190 | |
191 | if (!do_swap_account) |
192 | return; |
193 | |
194 | mutex_lock(&swap_cgroup_mutex); |
195 | ctrl = &swap_cgroup_ctrl[type]; |
196 | map = ctrl->map; |
197 | length = ctrl->length; |
198 | ctrl->map = NULL; |
199 | ctrl->length = 0; |
200 | mutex_unlock(&swap_cgroup_mutex); |
201 | |
202 | if (map) { |
203 | for (i = 0; i < length; i++) { |
204 | struct page *page = map[i]; |
205 | if (page) |
206 | __free_page(page); |
207 | if (!(i % SWAP_CLUSTER_MAX)) |
208 | cond_resched(); |
209 | } |
210 | vfree(map); |
211 | } |
212 | } |
213 |