diff options
Diffstat (limited to 'fs/file.c')
| -rw-r--r-- | fs/file.c | 255 |
1 files changed, 84 insertions, 171 deletions
| @@ -32,46 +32,28 @@ struct fdtable_defer { | |||
| 32 | */ | 32 | */ |
| 33 | static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list); | 33 | static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list); |
| 34 | 34 | ||
| 35 | 35 | static inline void * alloc_fdmem(unsigned int size) | |
| 36 | /* | ||
| 37 | * Allocate an fd array, using kmalloc or vmalloc. | ||
| 38 | * Note: the array isn't cleared at allocation time. | ||
| 39 | */ | ||
| 40 | struct file ** alloc_fd_array(int num) | ||
| 41 | { | 36 | { |
| 42 | struct file **new_fds; | ||
| 43 | int size = num * sizeof(struct file *); | ||
| 44 | |||
| 45 | if (size <= PAGE_SIZE) | 37 | if (size <= PAGE_SIZE) |
| 46 | new_fds = (struct file **) kmalloc(size, GFP_KERNEL); | 38 | return kmalloc(size, GFP_KERNEL); |
| 47 | else | 39 | else |
| 48 | new_fds = (struct file **) vmalloc(size); | 40 | return vmalloc(size); |
| 49 | return new_fds; | ||
| 50 | } | 41 | } |
| 51 | 42 | ||
| 52 | void free_fd_array(struct file **array, int num) | 43 | static inline void free_fdarr(struct fdtable *fdt) |
| 53 | { | 44 | { |
| 54 | int size = num * sizeof(struct file *); | 45 | if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) |
| 55 | 46 | kfree(fdt->fd); | |
| 56 | if (!array) { | ||
| 57 | printk (KERN_ERR "free_fd_array: array = 0 (num = %d)\n", num); | ||
| 58 | return; | ||
| 59 | } | ||
| 60 | |||
| 61 | if (num <= NR_OPEN_DEFAULT) /* Don't free the embedded fd array! */ | ||
| 62 | return; | ||
| 63 | else if (size <= PAGE_SIZE) | ||
| 64 | kfree(array); | ||
| 65 | else | 47 | else |
| 66 | vfree(array); | 48 | vfree(fdt->fd); |
| 67 | } | 49 | } |
| 68 | 50 | ||
| 69 | static void __free_fdtable(struct fdtable *fdt) | 51 | static inline void free_fdset(struct fdtable *fdt) |
| 70 | { | 52 | { |
| 71 | free_fdset(fdt->open_fds, fdt->max_fdset); | 53 | if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2)) |
| 72 | free_fdset(fdt->close_on_exec, fdt->max_fdset); | 54 | kfree(fdt->open_fds); |
| 73 | free_fd_array(fdt->fd, fdt->max_fds); | 55 | else |
| 74 | kfree(fdt); | 56 | vfree(fdt->open_fds); |
| 75 | } | 57 | } |
| 76 | 58 | ||
| 77 | static void free_fdtable_work(struct work_struct *work) | 59 | static void free_fdtable_work(struct work_struct *work) |
| @@ -86,41 +68,32 @@ static void free_fdtable_work(struct work_struct *work) | |||
| 86 | spin_unlock_bh(&f->lock); | 68 | spin_unlock_bh(&f->lock); |
| 87 | while(fdt) { | 69 | while(fdt) { |
| 88 | struct fdtable *next = fdt->next; | 70 | struct fdtable *next = fdt->next; |
| 89 | __free_fdtable(fdt); | 71 | vfree(fdt->fd); |
| 72 | free_fdset(fdt); | ||
| 73 | kfree(fdt); | ||
| 90 | fdt = next; | 74 | fdt = next; |
| 91 | } | 75 | } |
| 92 | } | 76 | } |
| 93 | 77 | ||
| 94 | static void free_fdtable_rcu(struct rcu_head *rcu) | 78 | void free_fdtable_rcu(struct rcu_head *rcu) |
| 95 | { | 79 | { |
| 96 | struct fdtable *fdt = container_of(rcu, struct fdtable, rcu); | 80 | struct fdtable *fdt = container_of(rcu, struct fdtable, rcu); |
| 97 | int fdset_size, fdarray_size; | ||
| 98 | struct fdtable_defer *fddef; | 81 | struct fdtable_defer *fddef; |
| 99 | 82 | ||
| 100 | BUG_ON(!fdt); | 83 | BUG_ON(!fdt); |
| 101 | fdset_size = fdt->max_fdset / 8; | ||
| 102 | fdarray_size = fdt->max_fds * sizeof(struct file *); | ||
| 103 | 84 | ||
| 104 | if (fdt->free_files) { | 85 | if (fdt->max_fds <= NR_OPEN_DEFAULT) { |
| 105 | /* | 86 | /* |
| 106 | * The this fdtable was embedded in the files structure | 87 | * This fdtable is embedded in the files structure and that |
| 107 | * and the files structure itself was getting destroyed. | 88 | * structure itself is getting destroyed. |
| 108 | * It is now safe to free the files structure. | ||
| 109 | */ | 89 | */ |
| 110 | kmem_cache_free(files_cachep, fdt->free_files); | 90 | kmem_cache_free(files_cachep, |
| 91 | container_of(fdt, struct files_struct, fdtab)); | ||
| 111 | return; | 92 | return; |
| 112 | } | 93 | } |
| 113 | if (fdt->max_fdset <= EMBEDDED_FD_SET_SIZE && | 94 | if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) { |
| 114 | fdt->max_fds <= NR_OPEN_DEFAULT) { | ||
| 115 | /* | ||
| 116 | * The fdtable was embedded | ||
| 117 | */ | ||
| 118 | return; | ||
| 119 | } | ||
| 120 | if (fdset_size <= PAGE_SIZE && fdarray_size <= PAGE_SIZE) { | ||
| 121 | kfree(fdt->open_fds); | ||
| 122 | kfree(fdt->close_on_exec); | ||
| 123 | kfree(fdt->fd); | 95 | kfree(fdt->fd); |
| 96 | kfree(fdt->open_fds); | ||
| 124 | kfree(fdt); | 97 | kfree(fdt); |
| 125 | } else { | 98 | } else { |
| 126 | fddef = &get_cpu_var(fdtable_defer_list); | 99 | fddef = &get_cpu_var(fdtable_defer_list); |
| @@ -134,136 +107,74 @@ static void free_fdtable_rcu(struct rcu_head *rcu) | |||
| 134 | } | 107 | } |
| 135 | } | 108 | } |
| 136 | 109 | ||
| 137 | void free_fdtable(struct fdtable *fdt) | ||
| 138 | { | ||
| 139 | if (fdt->free_files || | ||
| 140 | fdt->max_fdset > EMBEDDED_FD_SET_SIZE || | ||
| 141 | fdt->max_fds > NR_OPEN_DEFAULT) | ||
| 142 | call_rcu(&fdt->rcu, free_fdtable_rcu); | ||
| 143 | } | ||
| 144 | |||
| 145 | /* | 110 | /* |
| 146 | * Expand the fdset in the files_struct. Called with the files spinlock | 111 | * Expand the fdset in the files_struct. Called with the files spinlock |
| 147 | * held for write. | 112 | * held for write. |
| 148 | */ | 113 | */ |
| 149 | static void copy_fdtable(struct fdtable *nfdt, struct fdtable *fdt) | 114 | static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt) |
| 150 | { | 115 | { |
| 151 | int i; | 116 | unsigned int cpy, set; |
| 152 | int count; | ||
| 153 | |||
| 154 | BUG_ON(nfdt->max_fdset < fdt->max_fdset); | ||
| 155 | BUG_ON(nfdt->max_fds < fdt->max_fds); | ||
| 156 | /* Copy the existing tables and install the new pointers */ | ||
| 157 | |||
| 158 | i = fdt->max_fdset / (sizeof(unsigned long) * 8); | ||
| 159 | count = (nfdt->max_fdset - fdt->max_fdset) / 8; | ||
| 160 | 117 | ||
| 161 | /* | 118 | BUG_ON(nfdt->max_fds < ofdt->max_fds); |
| 162 | * Don't copy the entire array if the current fdset is | 119 | if (ofdt->max_fds == 0) |
| 163 | * not yet initialised. | 120 | return; |
| 164 | */ | ||
| 165 | if (i) { | ||
| 166 | memcpy (nfdt->open_fds, fdt->open_fds, | ||
| 167 | fdt->max_fdset/8); | ||
| 168 | memcpy (nfdt->close_on_exec, fdt->close_on_exec, | ||
| 169 | fdt->max_fdset/8); | ||
| 170 | memset (&nfdt->open_fds->fds_bits[i], 0, count); | ||
| 171 | memset (&nfdt->close_on_exec->fds_bits[i], 0, count); | ||
| 172 | } | ||
| 173 | |||
| 174 | /* Don't copy/clear the array if we are creating a new | ||
| 175 | fd array for fork() */ | ||
| 176 | if (fdt->max_fds) { | ||
| 177 | memcpy(nfdt->fd, fdt->fd, | ||
| 178 | fdt->max_fds * sizeof(struct file *)); | ||
| 179 | /* clear the remainder of the array */ | ||
| 180 | memset(&nfdt->fd[fdt->max_fds], 0, | ||
| 181 | (nfdt->max_fds - fdt->max_fds) * | ||
| 182 | sizeof(struct file *)); | ||
| 183 | } | ||
| 184 | } | ||
| 185 | |||
| 186 | /* | ||
| 187 | * Allocate an fdset array, using kmalloc or vmalloc. | ||
| 188 | * Note: the array isn't cleared at allocation time. | ||
| 189 | */ | ||
| 190 | fd_set * alloc_fdset(int num) | ||
| 191 | { | ||
| 192 | fd_set *new_fdset; | ||
| 193 | int size = num / 8; | ||
| 194 | 121 | ||
| 195 | if (size <= PAGE_SIZE) | 122 | cpy = ofdt->max_fds * sizeof(struct file *); |
| 196 | new_fdset = (fd_set *) kmalloc(size, GFP_KERNEL); | 123 | set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *); |
| 197 | else | 124 | memcpy(nfdt->fd, ofdt->fd, cpy); |
| 198 | new_fdset = (fd_set *) vmalloc(size); | 125 | memset((char *)(nfdt->fd) + cpy, 0, set); |
| 199 | return new_fdset; | 126 | |
| 127 | cpy = ofdt->max_fds / BITS_PER_BYTE; | ||
| 128 | set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE; | ||
| 129 | memcpy(nfdt->open_fds, ofdt->open_fds, cpy); | ||
| 130 | memset((char *)(nfdt->open_fds) + cpy, 0, set); | ||
| 131 | memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy); | ||
| 132 | memset((char *)(nfdt->close_on_exec) + cpy, 0, set); | ||
| 200 | } | 133 | } |
| 201 | 134 | ||
| 202 | void free_fdset(fd_set *array, int num) | 135 | static struct fdtable * alloc_fdtable(unsigned int nr) |
| 203 | { | 136 | { |
| 204 | if (num <= EMBEDDED_FD_SET_SIZE) /* Don't free an embedded fdset */ | 137 | struct fdtable *fdt; |
| 205 | return; | 138 | char *data; |
| 206 | else if (num <= 8 * PAGE_SIZE) | ||
| 207 | kfree(array); | ||
| 208 | else | ||
| 209 | vfree(array); | ||
| 210 | } | ||
| 211 | 139 | ||
| 212 | static struct fdtable *alloc_fdtable(int nr) | 140 | /* |
| 213 | { | 141 | * Figure out how many fds we actually want to support in this fdtable. |
| 214 | struct fdtable *fdt = NULL; | 142 | * Allocation steps are keyed to the size of the fdarray, since it |
| 215 | int nfds = 0; | 143 | * grows far faster than any of the other dynamic data. We try to fit |
| 216 | fd_set *new_openset = NULL, *new_execset = NULL; | 144 | * the fdarray into comfortable page-tuned chunks: starting at 1024B |
| 217 | struct file **new_fds; | 145 | * and growing in powers of two from there on. |
| 146 | */ | ||
| 147 | nr /= (1024 / sizeof(struct file *)); | ||
| 148 | nr = roundup_pow_of_two(nr + 1); | ||
| 149 | nr *= (1024 / sizeof(struct file *)); | ||
| 150 | if (nr > NR_OPEN) | ||
| 151 | nr = NR_OPEN; | ||
| 218 | 152 | ||
| 219 | fdt = kzalloc(sizeof(*fdt), GFP_KERNEL); | 153 | fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL); |
| 220 | if (!fdt) | 154 | if (!fdt) |
| 221 | goto out; | 155 | goto out; |
| 222 | 156 | fdt->max_fds = nr; | |
| 223 | nfds = max_t(int, 8 * L1_CACHE_BYTES, roundup_pow_of_two(nr + 1)); | 157 | data = alloc_fdmem(nr * sizeof(struct file *)); |
| 224 | if (nfds > NR_OPEN) | 158 | if (!data) |
| 225 | nfds = NR_OPEN; | 159 | goto out_fdt; |
| 160 | fdt->fd = (struct file **)data; | ||
| 161 | data = alloc_fdmem(max_t(unsigned int, | ||
| 162 | 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES)); | ||
| 163 | if (!data) | ||
| 164 | goto out_arr; | ||
| 165 | fdt->open_fds = (fd_set *)data; | ||
| 166 | data += nr / BITS_PER_BYTE; | ||
| 167 | fdt->close_on_exec = (fd_set *)data; | ||
| 168 | INIT_RCU_HEAD(&fdt->rcu); | ||
| 169 | fdt->next = NULL; | ||
| 226 | 170 | ||
| 227 | new_openset = alloc_fdset(nfds); | ||
| 228 | new_execset = alloc_fdset(nfds); | ||
| 229 | if (!new_openset || !new_execset) | ||
| 230 | goto out; | ||
| 231 | fdt->open_fds = new_openset; | ||
| 232 | fdt->close_on_exec = new_execset; | ||
| 233 | fdt->max_fdset = nfds; | ||
| 234 | |||
| 235 | nfds = NR_OPEN_DEFAULT; | ||
| 236 | /* | ||
| 237 | * Expand to the max in easy steps, and keep expanding it until | ||
| 238 | * we have enough for the requested fd array size. | ||
| 239 | */ | ||
| 240 | do { | ||
| 241 | #if NR_OPEN_DEFAULT < 256 | ||
| 242 | if (nfds < 256) | ||
| 243 | nfds = 256; | ||
| 244 | else | ||
| 245 | #endif | ||
| 246 | if (nfds < (PAGE_SIZE / sizeof(struct file *))) | ||
| 247 | nfds = PAGE_SIZE / sizeof(struct file *); | ||
| 248 | else { | ||
| 249 | nfds = nfds * 2; | ||
| 250 | if (nfds > NR_OPEN) | ||
| 251 | nfds = NR_OPEN; | ||
| 252 | } | ||
| 253 | } while (nfds <= nr); | ||
| 254 | new_fds = alloc_fd_array(nfds); | ||
| 255 | if (!new_fds) | ||
| 256 | goto out2; | ||
| 257 | fdt->fd = new_fds; | ||
| 258 | fdt->max_fds = nfds; | ||
| 259 | fdt->free_files = NULL; | ||
| 260 | return fdt; | 171 | return fdt; |
| 261 | out2: | 172 | |
| 262 | nfds = fdt->max_fdset; | 173 | out_arr: |
| 263 | out: | 174 | free_fdarr(fdt); |
| 264 | free_fdset(new_openset, nfds); | 175 | out_fdt: |
| 265 | free_fdset(new_execset, nfds); | ||
| 266 | kfree(fdt); | 176 | kfree(fdt); |
| 177 | out: | ||
| 267 | return NULL; | 178 | return NULL; |
| 268 | } | 179 | } |
| 269 | 180 | ||
| @@ -290,14 +201,17 @@ static int expand_fdtable(struct files_struct *files, int nr) | |||
| 290 | * we dropped the lock | 201 | * we dropped the lock |
| 291 | */ | 202 | */ |
| 292 | cur_fdt = files_fdtable(files); | 203 | cur_fdt = files_fdtable(files); |
| 293 | if (nr >= cur_fdt->max_fds || nr >= cur_fdt->max_fdset) { | 204 | if (nr >= cur_fdt->max_fds) { |
| 294 | /* Continue as planned */ | 205 | /* Continue as planned */ |
| 295 | copy_fdtable(new_fdt, cur_fdt); | 206 | copy_fdtable(new_fdt, cur_fdt); |
| 296 | rcu_assign_pointer(files->fdt, new_fdt); | 207 | rcu_assign_pointer(files->fdt, new_fdt); |
| 297 | free_fdtable(cur_fdt); | 208 | if (cur_fdt->max_fds > NR_OPEN_DEFAULT) |
| 209 | call_rcu(&cur_fdt->rcu, free_fdtable_rcu); | ||
| 298 | } else { | 210 | } else { |
| 299 | /* Somebody else expanded, so undo our attempt */ | 211 | /* Somebody else expanded, so undo our attempt */ |
| 300 | __free_fdtable(new_fdt); | 212 | free_fdarr(new_fdt); |
| 213 | free_fdset(new_fdt); | ||
| 214 | kfree(new_fdt); | ||
| 301 | } | 215 | } |
| 302 | return 1; | 216 | return 1; |
| 303 | } | 217 | } |
| @@ -316,11 +230,10 @@ int expand_files(struct files_struct *files, int nr) | |||
| 316 | 230 | ||
| 317 | fdt = files_fdtable(files); | 231 | fdt = files_fdtable(files); |
| 318 | /* Do we need to expand? */ | 232 | /* Do we need to expand? */ |
| 319 | if (nr < fdt->max_fdset && nr < fdt->max_fds) | 233 | if (nr < fdt->max_fds) |
| 320 | return 0; | 234 | return 0; |
| 321 | /* Can we expand? */ | 235 | /* Can we expand? */ |
| 322 | if (fdt->max_fdset >= NR_OPEN || fdt->max_fds >= NR_OPEN || | 236 | if (nr >= NR_OPEN) |
| 323 | nr >= NR_OPEN) | ||
| 324 | return -EMFILE; | 237 | return -EMFILE; |
| 325 | 238 | ||
| 326 | /* All good, so we try */ | 239 | /* All good, so we try */ |
