| 1 |
199 |
simons |
/*
|
| 2 |
|
|
* linux/fs/inode.c
|
| 3 |
|
|
*
|
| 4 |
|
|
* Copyright (C) 1991, 1992 Linus Torvalds
|
| 5 |
|
|
*
|
| 6 |
|
|
*/
|
| 7 |
|
|
|
| 8 |
|
|
/*
|
| 9 |
|
|
* uClinux revisions for NO_MM
|
| 10 |
|
|
* Copyright (C) 1998 Kenneth Albanowski <kjahds@kjahds.com>,
|
| 11 |
|
|
* The Silver Hammer Group, Ltd.
|
| 12 |
|
|
*/
|
| 13 |
|
|
|
| 14 |
|
|
#include <linux/stat.h>
|
| 15 |
|
|
#include <linux/sched.h>
|
| 16 |
|
|
#include <linux/kernel.h>
|
| 17 |
|
|
#include <linux/mm.h>
|
| 18 |
|
|
#include <linux/string.h>
|
| 19 |
|
|
|
| 20 |
|
|
#include <asm/system.h>
|
| 21 |
|
|
|
| 22 |
|
|
#define NR_IHASH 512
|
| 23 |
|
|
|
| 24 |
|
|
/*
|
| 25 |
|
|
* Be VERY careful when you access the inode hash table. There
|
| 26 |
|
|
* are some rather scary race conditions you need to take care of:
|
| 27 |
|
|
* - P1 tries to open file "xx", calls "iget()" with the proper
|
| 28 |
|
|
* inode number, but blocks because it's not on the list.
|
| 29 |
|
|
* - P2 deletes file "xx", gets the inode (which P1 has just read,
|
| 30 |
|
|
* but P1 hasn't woken up to the fact yet)
|
| 31 |
|
|
* - P2 iput()'s the inode, which now has i_nlink = 0
|
| 32 |
|
|
* - P1 wakes up and has the inode, but now P2 has made that
|
| 33 |
|
|
* inode invalid (but P1 has no way of knowing that).
|
| 34 |
|
|
*
|
| 35 |
|
|
* The "updating" counter makes sure that when P1 blocks on the
|
| 36 |
|
|
* iget(), P2 can't delete the inode from under it because P2
|
| 37 |
|
|
* will wait until P1 has been able to update the inode usage
|
| 38 |
|
|
* count so that the inode will stay in use until everybody has
|
| 39 |
|
|
* closed it..
|
| 40 |
|
|
*/
|
| 41 |
|
|
static struct inode_hash_entry {
|
| 42 |
|
|
struct inode * inode;
|
| 43 |
|
|
int updating;
|
| 44 |
|
|
} hash_table[NR_IHASH];
|
| 45 |
|
|
|
| 46 |
|
|
static struct inode * first_inode;
|
| 47 |
|
|
static struct wait_queue * inode_wait = NULL;
|
| 48 |
|
|
/* Keep these next two contiguous in memory for sysctl.c */
|
| 49 |
|
|
int nr_inodes = 0, nr_free_inodes = 0;
|
| 50 |
|
|
int max_inodes = NR_INODE;
|
| 51 |
|
|
|
| 52 |
|
|
static inline int const hashfn(kdev_t dev, unsigned int i)
|
| 53 |
|
|
{
|
| 54 |
|
|
return (HASHDEV(dev) ^ i) % NR_IHASH;
|
| 55 |
|
|
}
|
| 56 |
|
|
|
| 57 |
|
|
static inline struct inode_hash_entry * const hash(kdev_t dev, int i)
|
| 58 |
|
|
{
|
| 59 |
|
|
return hash_table + hashfn(dev, i);
|
| 60 |
|
|
}
|
| 61 |
|
|
|
| 62 |
|
|
static inline void insert_inode_free(struct inode *inode)
|
| 63 |
|
|
{
|
| 64 |
|
|
struct inode * prev, * next = first_inode;
|
| 65 |
|
|
|
| 66 |
|
|
first_inode = inode;
|
| 67 |
|
|
prev = next->i_prev;
|
| 68 |
|
|
inode->i_next = next;
|
| 69 |
|
|
inode->i_prev = prev;
|
| 70 |
|
|
prev->i_next = inode;
|
| 71 |
|
|
next->i_prev = inode;
|
| 72 |
|
|
}
|
| 73 |
|
|
|
| 74 |
|
|
static inline void remove_inode_free(struct inode *inode)
|
| 75 |
|
|
{
|
| 76 |
|
|
if (first_inode == inode)
|
| 77 |
|
|
first_inode = first_inode->i_next;
|
| 78 |
|
|
if (inode->i_next)
|
| 79 |
|
|
inode->i_next->i_prev = inode->i_prev;
|
| 80 |
|
|
if (inode->i_prev)
|
| 81 |
|
|
inode->i_prev->i_next = inode->i_next;
|
| 82 |
|
|
inode->i_next = inode->i_prev = NULL;
|
| 83 |
|
|
}
|
| 84 |
|
|
|
| 85 |
|
|
void insert_inode_hash(struct inode *inode)
|
| 86 |
|
|
{
|
| 87 |
|
|
struct inode_hash_entry *h;
|
| 88 |
|
|
h = hash(inode->i_dev, inode->i_ino);
|
| 89 |
|
|
|
| 90 |
|
|
inode->i_hash_next = h->inode;
|
| 91 |
|
|
inode->i_hash_prev = NULL;
|
| 92 |
|
|
if (inode->i_hash_next)
|
| 93 |
|
|
inode->i_hash_next->i_hash_prev = inode;
|
| 94 |
|
|
h->inode = inode;
|
| 95 |
|
|
}
|
| 96 |
|
|
|
| 97 |
|
|
static inline void remove_inode_hash(struct inode *inode)
|
| 98 |
|
|
{
|
| 99 |
|
|
struct inode_hash_entry *h;
|
| 100 |
|
|
h = hash(inode->i_dev, inode->i_ino);
|
| 101 |
|
|
|
| 102 |
|
|
if (h->inode == inode)
|
| 103 |
|
|
h->inode = inode->i_hash_next;
|
| 104 |
|
|
if (inode->i_hash_next)
|
| 105 |
|
|
inode->i_hash_next->i_hash_prev = inode->i_hash_prev;
|
| 106 |
|
|
if (inode->i_hash_prev)
|
| 107 |
|
|
inode->i_hash_prev->i_hash_next = inode->i_hash_next;
|
| 108 |
|
|
inode->i_hash_prev = inode->i_hash_next = NULL;
|
| 109 |
|
|
}
|
| 110 |
|
|
|
| 111 |
|
|
static inline void put_last_free(struct inode *inode)
|
| 112 |
|
|
{
|
| 113 |
|
|
remove_inode_free(inode);
|
| 114 |
|
|
inode->i_prev = first_inode->i_prev;
|
| 115 |
|
|
inode->i_prev->i_next = inode;
|
| 116 |
|
|
inode->i_next = first_inode;
|
| 117 |
|
|
inode->i_next->i_prev = inode;
|
| 118 |
|
|
}
|
| 119 |
|
|
|
| 120 |
|
|
int grow_inodes(void)
|
| 121 |
|
|
{
|
| 122 |
|
|
struct inode * inode;
|
| 123 |
|
|
int i;
|
| 124 |
|
|
|
| 125 |
|
|
if (!(inode = (struct inode*) get_free_page(GFP_KERNEL)))
|
| 126 |
|
|
return -ENOMEM;
|
| 127 |
|
|
|
| 128 |
|
|
i=PAGE_SIZE / sizeof(struct inode);
|
| 129 |
|
|
nr_inodes += i;
|
| 130 |
|
|
nr_free_inodes += i;
|
| 131 |
|
|
|
| 132 |
|
|
if (!first_inode)
|
| 133 |
|
|
inode->i_next = inode->i_prev = first_inode = inode++, i--;
|
| 134 |
|
|
|
| 135 |
|
|
for ( ; i ; i-- )
|
| 136 |
|
|
insert_inode_free(inode++);
|
| 137 |
|
|
return 0;
|
| 138 |
|
|
}
|
| 139 |
|
|
|
| 140 |
|
|
unsigned long inode_init(unsigned long start, unsigned long end)
|
| 141 |
|
|
{
|
| 142 |
|
|
memset(hash_table, 0, sizeof(hash_table));
|
| 143 |
|
|
first_inode = NULL;
|
| 144 |
|
|
return start;
|
| 145 |
|
|
}
|
| 146 |
|
|
|
| 147 |
|
|
static void __wait_on_inode(struct inode *);
|
| 148 |
|
|
|
| 149 |
|
|
static inline void wait_on_inode(struct inode * inode)
|
| 150 |
|
|
{
|
| 151 |
|
|
if (inode->i_lock)
|
| 152 |
|
|
__wait_on_inode(inode);
|
| 153 |
|
|
}
|
| 154 |
|
|
|
| 155 |
|
|
static inline void lock_inode(struct inode * inode)
|
| 156 |
|
|
{
|
| 157 |
|
|
wait_on_inode(inode);
|
| 158 |
|
|
inode->i_lock = 1;
|
| 159 |
|
|
}
|
| 160 |
|
|
|
| 161 |
|
|
static inline void unlock_inode(struct inode * inode)
|
| 162 |
|
|
{
|
| 163 |
|
|
inode->i_lock = 0;
|
| 164 |
|
|
wake_up(&inode->i_wait);
|
| 165 |
|
|
}
|
| 166 |
|
|
|
| 167 |
|
|
/*
|
| 168 |
|
|
* Note that we don't want to disturb any wait-queues when we discard
|
| 169 |
|
|
* an inode.
|
| 170 |
|
|
*
|
| 171 |
|
|
* Argghh. Got bitten by a gcc problem with inlining: no way to tell
|
| 172 |
|
|
* the compiler that the inline asm function 'memset' changes 'inode'.
|
| 173 |
|
|
* I've been searching for the bug for days, and was getting desperate.
|
| 174 |
|
|
* Finally looked at the assembler output... Grrr.
|
| 175 |
|
|
*
|
| 176 |
|
|
* The solution is the weird use of 'volatile'. Ho humm. Have to report
|
| 177 |
|
|
* it to the gcc lists, and hope we can do this more cleanly some day..
|
| 178 |
|
|
*/
|
| 179 |
|
|
void clear_inode(struct inode * inode)
|
| 180 |
|
|
{
|
| 181 |
|
|
struct wait_queue * wait;
|
| 182 |
|
|
|
| 183 |
|
|
/*
|
| 184 |
|
|
* We can clear inodes either when a last deref to the inode
|
| 185 |
|
|
* causes it to be deleted (reference count==1), or when we want to
|
| 186 |
|
|
* reuse it (reference count==0). Any other count is an error.
|
| 187 |
|
|
*/
|
| 188 |
|
|
if (inode->i_count > 1)
|
| 189 |
|
|
panic ("clear_inode: Inode still has references");
|
| 190 |
|
|
|
| 191 |
|
|
/*
|
| 192 |
|
|
* We are about to zap this inode. This operation may block,
|
| 193 |
|
|
* and it's imperative that we don't allow another process to
|
| 194 |
|
|
* grab it before it is completely pulled down. The i_count
|
| 195 |
|
|
* will prevent reuse of the inode by get_empty_inode(), but the
|
| 196 |
|
|
* i_condemned flag will also prevent __iget() from finding the
|
| 197 |
|
|
* inode until it is completely dead.
|
| 198 |
|
|
*/
|
| 199 |
|
|
inode->i_condemned = 1;
|
| 200 |
|
|
inode->i_count++;
|
| 201 |
|
|
|
| 202 |
|
|
truncate_inode_pages(inode, 0);
|
| 203 |
|
|
wait_on_inode(inode);
|
| 204 |
|
|
if (IS_WRITABLE(inode)) {
|
| 205 |
|
|
if (inode->i_sb && inode->i_sb->dq_op)
|
| 206 |
|
|
inode->i_sb->dq_op->drop(inode);
|
| 207 |
|
|
}
|
| 208 |
|
|
remove_inode_hash(inode);
|
| 209 |
|
|
remove_inode_free(inode);
|
| 210 |
|
|
wait = ((volatile struct inode *) inode)->i_wait;
|
| 211 |
|
|
if (--inode->i_count)
|
| 212 |
|
|
nr_free_inodes++;
|
| 213 |
|
|
memset(inode,0,sizeof(*inode));
|
| 214 |
|
|
((volatile struct inode *) inode)->i_wait = wait;
|
| 215 |
|
|
insert_inode_free(inode);
|
| 216 |
|
|
/*
|
| 217 |
|
|
* The inode is now reusable again, and the condemned flag is
|
| 218 |
|
|
* clear. Wake up anybody who is waiting on the condemned flag.
|
| 219 |
|
|
*/
|
| 220 |
|
|
wake_up(&inode->i_wait);
|
| 221 |
|
|
}
|
| 222 |
|
|
|
| 223 |
|
|
int fs_may_mount(kdev_t dev)
|
| 224 |
|
|
{
|
| 225 |
|
|
struct inode * inode, * next;
|
| 226 |
|
|
int i;
|
| 227 |
|
|
|
| 228 |
|
|
next = first_inode;
|
| 229 |
|
|
for (i = nr_inodes ; i > 0 ; i--) {
|
| 230 |
|
|
inode = next;
|
| 231 |
|
|
next = inode->i_next; /* clear_inode() changes the queues.. */
|
| 232 |
|
|
if (inode->i_dev != dev)
|
| 233 |
|
|
continue;
|
| 234 |
|
|
if (inode->i_count || inode->i_dirt || inode->i_lock)
|
| 235 |
|
|
return 0;
|
| 236 |
|
|
clear_inode(inode);
|
| 237 |
|
|
}
|
| 238 |
|
|
return 1;
|
| 239 |
|
|
}
|
| 240 |
|
|
|
| 241 |
|
|
int fs_may_umount(kdev_t dev, struct inode * mount_root)
|
| 242 |
|
|
{
|
| 243 |
|
|
struct inode * inode;
|
| 244 |
|
|
int i;
|
| 245 |
|
|
|
| 246 |
|
|
inode = first_inode;
|
| 247 |
|
|
for (i=0 ; i < nr_inodes ; i++, inode = inode->i_next) {
|
| 248 |
|
|
if (inode->i_dev != dev || !inode->i_count)
|
| 249 |
|
|
continue;
|
| 250 |
|
|
if (inode == mount_root && inode->i_count ==
|
| 251 |
|
|
(inode->i_mount != inode ? 1 : 2))
|
| 252 |
|
|
continue;
|
| 253 |
|
|
return 0;
|
| 254 |
|
|
}
|
| 255 |
|
|
return 1;
|
| 256 |
|
|
}
|
| 257 |
|
|
|
| 258 |
|
|
int fs_may_remount_ro(kdev_t dev)
|
| 259 |
|
|
{
|
| 260 |
|
|
struct file * file;
|
| 261 |
|
|
int i;
|
| 262 |
|
|
|
| 263 |
|
|
/* Check that no files are currently opened for writing. */
|
| 264 |
|
|
for (file = first_file, i=0; i<nr_files; i++, file=file->f_next) {
|
| 265 |
|
|
if (!file->f_count || !file->f_inode ||
|
| 266 |
|
|
file->f_inode->i_dev != dev)
|
| 267 |
|
|
continue;
|
| 268 |
|
|
if (S_ISREG(file->f_inode->i_mode) && (file->f_mode & 2))
|
| 269 |
|
|
return 0;
|
| 270 |
|
|
}
|
| 271 |
|
|
return 1;
|
| 272 |
|
|
}
|
| 273 |
|
|
|
| 274 |
|
|
static void write_inode(struct inode * inode)
|
| 275 |
|
|
{
|
| 276 |
|
|
if (!inode->i_dirt)
|
| 277 |
|
|
return;
|
| 278 |
|
|
wait_on_inode(inode);
|
| 279 |
|
|
if (!inode->i_dirt)
|
| 280 |
|
|
return;
|
| 281 |
|
|
if (!inode->i_sb || !inode->i_sb->s_op || !inode->i_sb->s_op->write_inode) {
|
| 282 |
|
|
inode->i_dirt = 0;
|
| 283 |
|
|
return;
|
| 284 |
|
|
}
|
| 285 |
|
|
inode->i_lock = 1;
|
| 286 |
|
|
inode->i_sb->s_op->write_inode(inode);
|
| 287 |
|
|
unlock_inode(inode);
|
| 288 |
|
|
}
|
| 289 |
|
|
|
| 290 |
|
|
static inline void read_inode(struct inode * inode)
|
| 291 |
|
|
{
|
| 292 |
|
|
lock_inode(inode);
|
| 293 |
|
|
if (inode->i_sb && inode->i_sb->s_op && inode->i_sb->s_op->read_inode)
|
| 294 |
|
|
inode->i_sb->s_op->read_inode(inode);
|
| 295 |
|
|
unlock_inode(inode);
|
| 296 |
|
|
}
|
| 297 |
|
|
|
| 298 |
|
|
/* POSIX UID/GID verification for setting inode attributes */
|
| 299 |
|
|
int inode_change_ok(struct inode *inode, struct iattr *attr)
|
| 300 |
|
|
{
|
| 301 |
|
|
/*
|
| 302 |
|
|
* If force is set do it anyway.
|
| 303 |
|
|
*/
|
| 304 |
|
|
|
| 305 |
|
|
if (attr->ia_valid & ATTR_FORCE)
|
| 306 |
|
|
return 0;
|
| 307 |
|
|
|
| 308 |
|
|
/* Make sure a caller can chown */
|
| 309 |
|
|
if ((attr->ia_valid & ATTR_UID) &&
|
| 310 |
|
|
(current->fsuid != inode->i_uid ||
|
| 311 |
|
|
attr->ia_uid != inode->i_uid) && !fsuser())
|
| 312 |
|
|
return -EPERM;
|
| 313 |
|
|
|
| 314 |
|
|
/* Make sure caller can chgrp */
|
| 315 |
|
|
if ((attr->ia_valid & ATTR_GID) &&
|
| 316 |
|
|
(!in_group_p(attr->ia_gid) && attr->ia_gid != inode->i_gid) &&
|
| 317 |
|
|
!fsuser())
|
| 318 |
|
|
return -EPERM;
|
| 319 |
|
|
|
| 320 |
|
|
/* Make sure a caller can chmod */
|
| 321 |
|
|
if (attr->ia_valid & ATTR_MODE) {
|
| 322 |
|
|
if ((current->fsuid != inode->i_uid) && !fsuser())
|
| 323 |
|
|
return -EPERM;
|
| 324 |
|
|
/* Also check the setgid bit! */
|
| 325 |
|
|
if (!fsuser() && !in_group_p((attr->ia_valid & ATTR_GID) ? attr->ia_gid :
|
| 326 |
|
|
inode->i_gid))
|
| 327 |
|
|
attr->ia_mode &= ~S_ISGID;
|
| 328 |
|
|
}
|
| 329 |
|
|
|
| 330 |
|
|
/* Check for setting the inode time */
|
| 331 |
|
|
if ((attr->ia_valid & ATTR_ATIME_SET) &&
|
| 332 |
|
|
((current->fsuid != inode->i_uid) && !fsuser()))
|
| 333 |
|
|
return -EPERM;
|
| 334 |
|
|
if ((attr->ia_valid & ATTR_MTIME_SET) &&
|
| 335 |
|
|
((current->fsuid != inode->i_uid) && !fsuser()))
|
| 336 |
|
|
return -EPERM;
|
| 337 |
|
|
return 0;
|
| 338 |
|
|
}
|
| 339 |
|
|
|
| 340 |
|
|
/*
|
| 341 |
|
|
* Set the appropriate attributes from an attribute structure into
|
| 342 |
|
|
* the inode structure.
|
| 343 |
|
|
*/
|
| 344 |
|
|
void inode_setattr(struct inode *inode, struct iattr *attr)
|
| 345 |
|
|
{
|
| 346 |
|
|
if (attr->ia_valid & ATTR_UID)
|
| 347 |
|
|
inode->i_uid = attr->ia_uid;
|
| 348 |
|
|
if (attr->ia_valid & ATTR_GID)
|
| 349 |
|
|
inode->i_gid = attr->ia_gid;
|
| 350 |
|
|
if (attr->ia_valid & ATTR_SIZE)
|
| 351 |
|
|
inode->i_size = attr->ia_size;
|
| 352 |
|
|
if (attr->ia_valid & ATTR_ATIME)
|
| 353 |
|
|
inode->i_atime = attr->ia_atime;
|
| 354 |
|
|
if (attr->ia_valid & ATTR_MTIME)
|
| 355 |
|
|
inode->i_mtime = attr->ia_mtime;
|
| 356 |
|
|
if (attr->ia_valid & ATTR_CTIME)
|
| 357 |
|
|
inode->i_ctime = attr->ia_ctime;
|
| 358 |
|
|
if (attr->ia_valid & ATTR_MODE) {
|
| 359 |
|
|
inode->i_mode = attr->ia_mode;
|
| 360 |
|
|
if (!fsuser() && !in_group_p(inode->i_gid))
|
| 361 |
|
|
inode->i_mode &= ~S_ISGID;
|
| 362 |
|
|
}
|
| 363 |
|
|
inode->i_dirt = 1;
|
| 364 |
|
|
}
|
| 365 |
|
|
|
| 366 |
|
|
/*
|
| 367 |
|
|
* notify_change is called for inode-changing operations such as
|
| 368 |
|
|
* chown, chmod, utime, and truncate. It is guaranteed (unlike
|
| 369 |
|
|
* write_inode) to be called from the context of the user requesting
|
| 370 |
|
|
* the change.
|
| 371 |
|
|
*/
|
| 372 |
|
|
|
| 373 |
|
|
int notify_change(struct inode * inode, struct iattr *attr)
|
| 374 |
|
|
{
|
| 375 |
|
|
int retval;
|
| 376 |
|
|
|
| 377 |
|
|
attr->ia_ctime = CURRENT_TIME;
|
| 378 |
|
|
if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
|
| 379 |
|
|
if (!(attr->ia_valid & ATTR_ATIME_SET))
|
| 380 |
|
|
attr->ia_atime = attr->ia_ctime;
|
| 381 |
|
|
if (!(attr->ia_valid & ATTR_MTIME_SET))
|
| 382 |
|
|
attr->ia_mtime = attr->ia_ctime;
|
| 383 |
|
|
}
|
| 384 |
|
|
|
| 385 |
|
|
if (inode->i_sb && inode->i_sb->s_op &&
|
| 386 |
|
|
inode->i_sb->s_op->notify_change)
|
| 387 |
|
|
return inode->i_sb->s_op->notify_change(inode, attr);
|
| 388 |
|
|
|
| 389 |
|
|
if ((retval = inode_change_ok(inode, attr)) != 0)
|
| 390 |
|
|
return retval;
|
| 391 |
|
|
|
| 392 |
|
|
inode_setattr(inode, attr);
|
| 393 |
|
|
return 0;
|
| 394 |
|
|
}
|
| 395 |
|
|
|
| 396 |
|
|
/*
|
| 397 |
|
|
* bmap is needed for demand-loading and paging: if this function
|
| 398 |
|
|
* doesn't exist for a filesystem, then those things are impossible:
|
| 399 |
|
|
* executables cannot be run from the filesystem etc...
|
| 400 |
|
|
*
|
| 401 |
|
|
* This isn't as bad as it sounds: the read-routines might still work,
|
| 402 |
|
|
* so the filesystem would be otherwise ok (for example, you might have
|
| 403 |
|
|
* a DOS filesystem, which doesn't lend itself to bmap very well, but
|
| 404 |
|
|
* you could still transfer files to/from the filesystem)
|
| 405 |
|
|
*/
|
| 406 |
|
|
int bmap(struct inode * inode, int block)
|
| 407 |
|
|
{
|
| 408 |
|
|
if (inode->i_op && inode->i_op->bmap)
|
| 409 |
|
|
return inode->i_op->bmap(inode,block);
|
| 410 |
|
|
return 0;
|
| 411 |
|
|
}
|
| 412 |
|
|
|
| 413 |
|
|
void invalidate_inodes(kdev_t dev)
|
| 414 |
|
|
{
|
| 415 |
|
|
struct inode * inode, * next;
|
| 416 |
|
|
int i;
|
| 417 |
|
|
|
| 418 |
|
|
next = first_inode;
|
| 419 |
|
|
for(i = nr_inodes ; i > 0 ; i--) {
|
| 420 |
|
|
inode = next;
|
| 421 |
|
|
next = inode->i_next; /* clear_inode() changes the queues.. */
|
| 422 |
|
|
if (inode->i_dev != dev)
|
| 423 |
|
|
continue;
|
| 424 |
|
|
if (inode->i_count || inode->i_dirt || inode->i_lock) {
|
| 425 |
|
|
printk("VFS: inode busy on removed device %s\n",
|
| 426 |
|
|
kdevname(dev));
|
| 427 |
|
|
continue;
|
| 428 |
|
|
}
|
| 429 |
|
|
clear_inode(inode);
|
| 430 |
|
|
}
|
| 431 |
|
|
}
|
| 432 |
|
|
|
| 433 |
|
|
void sync_inodes(kdev_t dev)
|
| 434 |
|
|
{
|
| 435 |
|
|
int i;
|
| 436 |
|
|
struct inode * inode;
|
| 437 |
|
|
|
| 438 |
|
|
inode = first_inode;
|
| 439 |
|
|
for(i = 0; i < nr_inodes*2; i++, inode = inode->i_next) {
|
| 440 |
|
|
if (dev && inode->i_dev != dev)
|
| 441 |
|
|
continue;
|
| 442 |
|
|
wait_on_inode(inode);
|
| 443 |
|
|
if (inode->i_dirt)
|
| 444 |
|
|
write_inode(inode);
|
| 445 |
|
|
}
|
| 446 |
|
|
}
|
| 447 |
|
|
|
| 448 |
|
|
void iput(struct inode * inode)
|
| 449 |
|
|
{
|
| 450 |
|
|
if (!inode)
|
| 451 |
|
|
return;
|
| 452 |
|
|
wait_on_inode(inode);
|
| 453 |
|
|
if (!inode->i_count) {
|
| 454 |
|
|
printk("VFS: iput: trying to free free inode\n");
|
| 455 |
|
|
printk("VFS: device %s, inode %lu, mode=0%07o\n",
|
| 456 |
|
|
kdevname(inode->i_rdev), inode->i_ino, inode->i_mode);
|
| 457 |
|
|
return;
|
| 458 |
|
|
}
|
| 459 |
|
|
if (inode->i_pipe)
|
| 460 |
|
|
wake_up_interruptible(&PIPE_WAIT(*inode));
|
| 461 |
|
|
|
| 462 |
|
|
repeat:
|
| 463 |
|
|
if (inode->i_count>1) {
|
| 464 |
|
|
inode->i_count--;
|
| 465 |
|
|
return;
|
| 466 |
|
|
}
|
| 467 |
|
|
|
| 468 |
|
|
wake_up(&inode_wait);
|
| 469 |
|
|
if (inode->i_pipe) {
|
| 470 |
|
|
unsigned long page = (unsigned long) PIPE_BASE(*inode);
|
| 471 |
|
|
PIPE_BASE(*inode) = NULL;
|
| 472 |
|
|
free_page(page);
|
| 473 |
|
|
}
|
| 474 |
|
|
|
| 475 |
|
|
if (inode->i_sb && inode->i_sb->s_op && inode->i_sb->s_op->put_inode) {
|
| 476 |
|
|
inode->i_sb->s_op->put_inode(inode);
|
| 477 |
|
|
if (!inode->i_nlink)
|
| 478 |
|
|
return;
|
| 479 |
|
|
}
|
| 480 |
|
|
|
| 481 |
|
|
if (inode->i_dirt) {
|
| 482 |
|
|
write_inode(inode); /* we can sleep - so do again */
|
| 483 |
|
|
wait_on_inode(inode);
|
| 484 |
|
|
goto repeat;
|
| 485 |
|
|
}
|
| 486 |
|
|
|
| 487 |
|
|
if (IS_WRITABLE(inode)) {
|
| 488 |
|
|
if (inode->i_sb && inode->i_sb->dq_op) {
|
| 489 |
|
|
/* Here we can sleep also. Let's do it again
|
| 490 |
|
|
* Dmitry Gorodchanin 02/11/96
|
| 491 |
|
|
*/
|
| 492 |
|
|
inode->i_lock = 1;
|
| 493 |
|
|
inode->i_sb->dq_op->drop(inode);
|
| 494 |
|
|
unlock_inode(inode);
|
| 495 |
|
|
goto repeat;
|
| 496 |
|
|
}
|
| 497 |
|
|
}
|
| 498 |
|
|
|
| 499 |
|
|
inode->i_count--;
|
| 500 |
|
|
|
| 501 |
|
|
if (inode->i_count)
|
| 502 |
|
|
/*
|
| 503 |
|
|
* Huoh, we were supposed to be the last user, but someone has
|
| 504 |
|
|
* grabbed it while we were sleeping. Dont destroy inode VM
|
| 505 |
|
|
* mappings, it might cause a memory leak.
|
| 506 |
|
|
*/
|
| 507 |
|
|
return;
|
| 508 |
|
|
|
| 509 |
|
|
#ifndef NO_MM
|
| 510 |
|
|
if (inode->i_mmap) {
|
| 511 |
|
|
printk("iput: inode %lu on device %s still has mappings.\n",
|
| 512 |
|
|
inode->i_ino, kdevname(inode->i_dev));
|
| 513 |
|
|
inode->i_mmap = NULL;
|
| 514 |
|
|
}
|
| 515 |
|
|
#endif /* !NO_MM */
|
| 516 |
|
|
|
| 517 |
|
|
nr_free_inodes++;
|
| 518 |
|
|
return;
|
| 519 |
|
|
}
|
| 520 |
|
|
|
| 521 |
|
|
struct inode * get_empty_inode(void)
|
| 522 |
|
|
{
|
| 523 |
|
|
static int ino = 0;
|
| 524 |
|
|
struct inode * inode, * best;
|
| 525 |
|
|
unsigned long badness;
|
| 526 |
|
|
int i;
|
| 527 |
|
|
|
| 528 |
|
|
if (nr_inodes < max_inodes && nr_free_inodes < (nr_inodes >> 1))
|
| 529 |
|
|
grow_inodes();
|
| 530 |
|
|
repeat:
|
| 531 |
|
|
inode = first_inode;
|
| 532 |
|
|
best = NULL;
|
| 533 |
|
|
badness = 1000;
|
| 534 |
|
|
for (i = nr_inodes/2; i > 0; i--,inode = inode->i_next) {
|
| 535 |
|
|
if (!inode->i_count) {
|
| 536 |
|
|
unsigned long i = 999;
|
| 537 |
|
|
if (!(inode->i_lock || inode->i_dirt))
|
| 538 |
|
|
i = inode->i_nrpages;
|
| 539 |
|
|
if (i < badness) {
|
| 540 |
|
|
best = inode;
|
| 541 |
|
|
if (!i)
|
| 542 |
|
|
goto found_good;
|
| 543 |
|
|
badness = i;
|
| 544 |
|
|
}
|
| 545 |
|
|
}
|
| 546 |
|
|
}
|
| 547 |
|
|
if (nr_inodes < max_inodes) {
|
| 548 |
|
|
if (grow_inodes() == 0)
|
| 549 |
|
|
goto repeat;
|
| 550 |
|
|
best = NULL;
|
| 551 |
|
|
}
|
| 552 |
|
|
if (!best) {
|
| 553 |
|
|
printk("VFS: No free inodes - contact Linus\n");
|
| 554 |
|
|
sleep_on(&inode_wait);
|
| 555 |
|
|
goto repeat;
|
| 556 |
|
|
}
|
| 557 |
|
|
if (best->i_lock) {
|
| 558 |
|
|
wait_on_inode(best);
|
| 559 |
|
|
goto repeat;
|
| 560 |
|
|
}
|
| 561 |
|
|
if (best->i_dirt) {
|
| 562 |
|
|
write_inode(best);
|
| 563 |
|
|
goto repeat;
|
| 564 |
|
|
}
|
| 565 |
|
|
if (best->i_count)
|
| 566 |
|
|
goto repeat;
|
| 567 |
|
|
found_good:
|
| 568 |
|
|
clear_inode(best);
|
| 569 |
|
|
best->i_count = 1;
|
| 570 |
|
|
best->i_nlink = 1;
|
| 571 |
|
|
best->i_version = ++event;
|
| 572 |
|
|
best->i_sem.count = 1;
|
| 573 |
|
|
best->i_ino = ++ino;
|
| 574 |
|
|
best->i_dev = 0;
|
| 575 |
|
|
nr_free_inodes--;
|
| 576 |
|
|
if (nr_free_inodes < 0) {
|
| 577 |
|
|
printk ("VFS: get_empty_inode: bad free inode count.\n");
|
| 578 |
|
|
nr_free_inodes = 0;
|
| 579 |
|
|
}
|
| 580 |
|
|
return best;
|
| 581 |
|
|
}
|
| 582 |
|
|
|
| 583 |
|
|
struct inode * get_pipe_inode(void)
|
| 584 |
|
|
{
|
| 585 |
|
|
struct inode * inode;
|
| 586 |
|
|
extern struct inode_operations pipe_inode_operations;
|
| 587 |
|
|
|
| 588 |
|
|
if (!(inode = get_empty_inode()))
|
| 589 |
|
|
return NULL;
|
| 590 |
|
|
if (!(PIPE_BASE(*inode) = (char*) __get_free_page(GFP_USER))) {
|
| 591 |
|
|
iput(inode);
|
| 592 |
|
|
return NULL;
|
| 593 |
|
|
}
|
| 594 |
|
|
inode->i_op = &pipe_inode_operations;
|
| 595 |
|
|
inode->i_count = 2; /* sum of readers/writers */
|
| 596 |
|
|
PIPE_WAIT(*inode) = NULL;
|
| 597 |
|
|
PIPE_START(*inode) = PIPE_LEN(*inode) = 0;
|
| 598 |
|
|
PIPE_RD_OPENERS(*inode) = PIPE_WR_OPENERS(*inode) = 0;
|
| 599 |
|
|
PIPE_READERS(*inode) = PIPE_WRITERS(*inode) = 1;
|
| 600 |
|
|
PIPE_LOCK(*inode) = 0;
|
| 601 |
|
|
inode->i_pipe = 1;
|
| 602 |
|
|
inode->i_mode |= S_IFIFO | S_IRUSR | S_IWUSR;
|
| 603 |
|
|
inode->i_uid = current->fsuid;
|
| 604 |
|
|
inode->i_gid = current->fsgid;
|
| 605 |
|
|
inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
|
| 606 |
|
|
inode->i_blksize = PAGE_SIZE;
|
| 607 |
|
|
return inode;
|
| 608 |
|
|
}
|
| 609 |
|
|
|
| 610 |
|
|
struct inode *__iget(struct super_block * sb, int nr, int crossmntp)
|
| 611 |
|
|
{
|
| 612 |
|
|
static struct wait_queue * update_wait = NULL;
|
| 613 |
|
|
struct inode_hash_entry * h;
|
| 614 |
|
|
struct inode * inode;
|
| 615 |
|
|
struct inode * empty = NULL;
|
| 616 |
|
|
|
| 617 |
|
|
if (!sb)
|
| 618 |
|
|
panic("VFS: iget with sb==NULL");
|
| 619 |
|
|
h = hash(sb->s_dev, nr);
|
| 620 |
|
|
repeat:
|
| 621 |
|
|
for (inode = h->inode; inode ; inode = inode->i_hash_next)
|
| 622 |
|
|
if (inode->i_dev == sb->s_dev && inode->i_ino == nr)
|
| 623 |
|
|
goto found_it;
|
| 624 |
|
|
if (!empty) {
|
| 625 |
|
|
/*
|
| 626 |
|
|
* If we sleep here before we have found an inode
|
| 627 |
|
|
* we need to make sure nobody does anything bad
|
| 628 |
|
|
* to the inode while we sleep, because otherwise
|
| 629 |
|
|
* we may return an inode that is not valid any
|
| 630 |
|
|
* more when we wake up..
|
| 631 |
|
|
*/
|
| 632 |
|
|
h->updating++;
|
| 633 |
|
|
empty = get_empty_inode();
|
| 634 |
|
|
if (!--h->updating)
|
| 635 |
|
|
wake_up(&update_wait);
|
| 636 |
|
|
if (empty)
|
| 637 |
|
|
goto repeat;
|
| 638 |
|
|
return (NULL);
|
| 639 |
|
|
}
|
| 640 |
|
|
inode = empty;
|
| 641 |
|
|
inode->i_sb = sb;
|
| 642 |
|
|
inode->i_dev = sb->s_dev;
|
| 643 |
|
|
inode->i_ino = nr;
|
| 644 |
|
|
inode->i_flags = sb->s_flags;
|
| 645 |
|
|
put_last_free(inode);
|
| 646 |
|
|
insert_inode_hash(inode);
|
| 647 |
|
|
read_inode(inode);
|
| 648 |
|
|
goto return_it;
|
| 649 |
|
|
|
| 650 |
|
|
found_it:
|
| 651 |
|
|
/*
|
| 652 |
|
|
* The inode may currently be being pulled down by
|
| 653 |
|
|
* clear_inode(). Avoid it if so. If we get past this, then
|
| 654 |
|
|
* the increment of i_count will prevent the inode's reuse.
|
| 655 |
|
|
*/
|
| 656 |
|
|
if (inode->i_condemned) {
|
| 657 |
|
|
sleep_on(&inode->i_wait);
|
| 658 |
|
|
goto repeat;
|
| 659 |
|
|
}
|
| 660 |
|
|
if (!inode->i_count)
|
| 661 |
|
|
nr_free_inodes--;
|
| 662 |
|
|
inode->i_count++;
|
| 663 |
|
|
wait_on_inode(inode);
|
| 664 |
|
|
if (inode->i_dev != sb->s_dev || inode->i_ino != nr) {
|
| 665 |
|
|
printk("Whee.. inode changed from under us. Tell Linus\n");
|
| 666 |
|
|
iput(inode);
|
| 667 |
|
|
goto repeat;
|
| 668 |
|
|
}
|
| 669 |
|
|
if (crossmntp && inode->i_mount) {
|
| 670 |
|
|
struct inode * tmp = inode->i_mount;
|
| 671 |
|
|
tmp->i_count++;
|
| 672 |
|
|
iput(inode);
|
| 673 |
|
|
inode = tmp;
|
| 674 |
|
|
wait_on_inode(inode);
|
| 675 |
|
|
}
|
| 676 |
|
|
if (empty)
|
| 677 |
|
|
iput(empty);
|
| 678 |
|
|
|
| 679 |
|
|
return_it:
|
| 680 |
|
|
while (h->updating)
|
| 681 |
|
|
sleep_on(&update_wait);
|
| 682 |
|
|
return inode;
|
| 683 |
|
|
}
|
| 684 |
|
|
|
| 685 |
|
|
/*
|
| 686 |
|
|
* The "new" scheduling primitives (new as of 0.97 or so) allow this to
|
| 687 |
|
|
* be done without disabling interrupts (other than in the actual queue
|
| 688 |
|
|
* updating things: only a couple of 386 instructions). This should be
|
| 689 |
|
|
* much better for interrupt latency.
|
| 690 |
|
|
*/
|
| 691 |
|
|
static void __wait_on_inode(struct inode * inode)
|
| 692 |
|
|
{
|
| 693 |
|
|
struct wait_queue wait = { current, NULL };
|
| 694 |
|
|
|
| 695 |
|
|
add_wait_queue(&inode->i_wait, &wait);
|
| 696 |
|
|
repeat:
|
| 697 |
|
|
current->state = TASK_UNINTERRUPTIBLE;
|
| 698 |
|
|
if (inode->i_lock) {
|
| 699 |
|
|
schedule();
|
| 700 |
|
|
goto repeat;
|
| 701 |
|
|
}
|
| 702 |
|
|
remove_wait_queue(&inode->i_wait, &wait);
|
| 703 |
|
|
current->state = TASK_RUNNING;
|
| 704 |
|
|
}
|
