diff options
author | Eric Paris <eparis@redhat.com> | 2009-12-17 21:24:27 -0500 |
---|---|---|
committer | Eric Paris <eparis@redhat.com> | 2010-07-28 09:58:57 -0400 |
commit | 5444e2981c31d0ed7465475e451b8437084337e5 (patch) | |
tree | 66b6d84b7aab886b44a3467a139d258d9aba09df /fs/notify/inode_mark.c | |
parent | 32c3263221bd63316815286dccacdc7abfd7f3c4 (diff) |
fsnotify: split generic and inode specific mark code
currently all marking is done by functions in inode-mark.c. Some of this
is pretty generic and should be instead done in a generic function and we
should only put the inode specific code in inode-mark.c
Signed-off-by: Eric Paris <eparis@redhat.com>
Diffstat (limited to 'fs/notify/inode_mark.c')
-rw-r--r-- | fs/notify/inode_mark.c | 246 |
1 files changed, 19 insertions, 227 deletions
diff --git a/fs/notify/inode_mark.c b/fs/notify/inode_mark.c index ba6f9833561b..c925579ba011 100644 --- a/fs/notify/inode_mark.c +++ b/fs/notify/inode_mark.c | |||
@@ -16,72 +16,6 @@ | |||
16 | * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | 16 | * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. |
17 | */ | 17 | */ |
18 | 18 | ||
19 | /* | ||
20 | * fsnotify inode mark locking/lifetime/and refcnting | ||
21 | * | ||
22 | * REFCNT: | ||
23 | * The mark->refcnt tells how many "things" in the kernel currently are | ||
24 | * referencing this object. The object typically will live inside the kernel | ||
25 | * with a refcnt of 2, one for each list it is on (i_list, g_list). Any task | ||
26 | * which can find this object holding the appropriete locks, can take a reference | ||
27 | * and the object itself is guarenteed to survive until the reference is dropped. | ||
28 | * | ||
29 | * LOCKING: | ||
30 | * There are 3 spinlocks involved with fsnotify inode marks and they MUST | ||
31 | * be taken in order as follows: | ||
32 | * | ||
33 | * mark->lock | ||
34 | * group->mark_lock | ||
35 | * inode->i_lock | ||
36 | * | ||
37 | * mark->lock protects 2 things, mark->group and mark->inode. You must hold | ||
38 | * that lock to dereference either of these things (they could be NULL even with | ||
39 | * the lock) | ||
40 | * | ||
41 | * group->mark_lock protects the marks_list anchored inside a given group | ||
42 | * and each mark is hooked via the g_list. It also sorta protects the | ||
43 | * free_g_list, which when used is anchored by a private list on the stack of the | ||
44 | * task which held the group->mark_lock. | ||
45 | * | ||
46 | * inode->i_lock protects the i_fsnotify_marks list anchored inside a | ||
47 | * given inode and each mark is hooked via the i_list. (and sorta the | ||
48 | * free_i_list) | ||
49 | * | ||
50 | * | ||
51 | * LIFETIME: | ||
52 | * Inode marks survive between when they are added to an inode and when their | ||
53 | * refcnt==0. | ||
54 | * | ||
55 | * The inode mark can be cleared for a number of different reasons including: | ||
56 | * - The inode is unlinked for the last time. (fsnotify_inode_remove) | ||
57 | * - The inode is being evicted from cache. (fsnotify_inode_delete) | ||
58 | * - The fs the inode is on is unmounted. (fsnotify_inode_delete/fsnotify_unmount_inodes) | ||
59 | * - Something explicitly requests that it be removed. (fsnotify_destroy_mark) | ||
60 | * - The fsnotify_group associated with the mark is going away and all such marks | ||
61 | * need to be cleaned up. (fsnotify_clear_marks_by_group) | ||
62 | * | ||
63 | * Worst case we are given an inode and need to clean up all the marks on that | ||
64 | * inode. We take i_lock and walk the i_fsnotify_marks safely. For each | ||
65 | * mark on the list we take a reference (so the mark can't disappear under us). | ||
66 | * We remove that mark form the inode's list of marks and we add this mark to a | ||
67 | * private list anchored on the stack using i_free_list; At this point we no | ||
68 | * longer fear anything finding the mark using the inode's list of marks. | ||
69 | * | ||
70 | * We can safely and locklessly run the private list on the stack of everything | ||
71 | * we just unattached from the original inode. For each mark on the private list | ||
72 | * we grab the mark-> and can thus dereference mark->group and mark->inode. If | ||
73 | * we see the group and inode are not NULL we take those locks. Now holding all | ||
74 | * 3 locks we can completely remove the mark from other tasks finding it in the | ||
75 | * future. Remember, 10 things might already be referencing this mark, but they | ||
76 | * better be holding a ref. We drop our reference we took before we unhooked it | ||
77 | * from the inode. When the ref hits 0 we can free the mark. | ||
78 | * | ||
79 | * Very similarly for freeing by group, except we use free_g_list. | ||
80 | * | ||
81 | * This has the very interesting property of being able to run concurrently with | ||
82 | * any (or all) other directions. | ||
83 | */ | ||
84 | |||
85 | #include <linux/fs.h> | 19 | #include <linux/fs.h> |
86 | #include <linux/init.h> | 20 | #include <linux/init.h> |
87 | #include <linux/kernel.h> | 21 | #include <linux/kernel.h> |
@@ -95,17 +29,6 @@ | |||
95 | #include <linux/fsnotify_backend.h> | 29 | #include <linux/fsnotify_backend.h> |
96 | #include "fsnotify.h" | 30 | #include "fsnotify.h" |
97 | 31 | ||
98 | void fsnotify_get_mark(struct fsnotify_mark *mark) | ||
99 | { | ||
100 | atomic_inc(&mark->refcnt); | ||
101 | } | ||
102 | |||
103 | void fsnotify_put_mark(struct fsnotify_mark *mark) | ||
104 | { | ||
105 | if (atomic_dec_and_test(&mark->refcnt)) | ||
106 | mark->free_mark(mark); | ||
107 | } | ||
108 | |||
109 | /* | 32 | /* |
110 | * Recalculate the mask of events relevant to a given inode locked. | 33 | * Recalculate the mask of events relevant to a given inode locked. |
111 | */ | 34 | */ |
@@ -135,44 +58,18 @@ void fsnotify_recalc_inode_mask(struct inode *inode) | |||
135 | __fsnotify_update_child_dentry_flags(inode); | 58 | __fsnotify_update_child_dentry_flags(inode); |
136 | } | 59 | } |
137 | 60 | ||
138 | /* | 61 | void fsnotify_destroy_inode_mark(struct fsnotify_mark *mark) |
139 | * Any time a mark is getting freed we end up here. | ||
140 | * The caller had better be holding a reference to this mark so we don't actually | ||
141 | * do the final put under the mark->lock | ||
142 | */ | ||
143 | void fsnotify_destroy_mark(struct fsnotify_mark *mark) | ||
144 | { | 62 | { |
145 | struct fsnotify_group *group; | 63 | struct inode *inode = mark->i.inode; |
146 | struct inode *inode; | ||
147 | |||
148 | spin_lock(&mark->lock); | ||
149 | 64 | ||
150 | group = mark->group; | 65 | assert_spin_locked(&mark->lock); |
151 | inode = mark->i.inode; | 66 | assert_spin_locked(&mark->group->mark_lock); |
152 | 67 | ||
153 | BUG_ON(group && !inode); | ||
154 | BUG_ON(!group && inode); | ||
155 | |||
156 | /* if !group something else already marked this to die */ | ||
157 | if (!group) { | ||
158 | spin_unlock(&mark->lock); | ||
159 | return; | ||
160 | } | ||
161 | |||
162 | /* 1 from caller and 1 for being on i_list/g_list */ | ||
163 | BUG_ON(atomic_read(&mark->refcnt) < 2); | ||
164 | |||
165 | spin_lock(&group->mark_lock); | ||
166 | spin_lock(&inode->i_lock); | 68 | spin_lock(&inode->i_lock); |
167 | 69 | ||
168 | hlist_del_init(&mark->i.i_list); | 70 | hlist_del_init(&mark->i.i_list); |
169 | mark->i.inode = NULL; | 71 | mark->i.inode = NULL; |
170 | 72 | ||
171 | list_del_init(&mark->g_list); | ||
172 | mark->group = NULL; | ||
173 | |||
174 | fsnotify_put_mark(mark); /* for i_list and g_list */ | ||
175 | |||
176 | /* | 73 | /* |
177 | * this mark is now off the inode->i_fsnotify_marks list and we | 74 | * this mark is now off the inode->i_fsnotify_marks list and we |
178 | * hold the inode->i_lock, so this is the perfect time to update the | 75 | * hold the inode->i_lock, so this is the perfect time to update the |
@@ -181,61 +78,6 @@ void fsnotify_destroy_mark(struct fsnotify_mark *mark) | |||
181 | fsnotify_recalc_inode_mask_locked(inode); | 78 | fsnotify_recalc_inode_mask_locked(inode); |
182 | 79 | ||
183 | spin_unlock(&inode->i_lock); | 80 | spin_unlock(&inode->i_lock); |
184 | spin_unlock(&group->mark_lock); | ||
185 | spin_unlock(&mark->lock); | ||
186 | |||
187 | /* | ||
188 | * Some groups like to know that marks are being freed. This is a | ||
189 | * callback to the group function to let it know that this mark | ||
190 | * is being freed. | ||
191 | */ | ||
192 | if (group->ops->freeing_mark) | ||
193 | group->ops->freeing_mark(mark, group); | ||
194 | |||
195 | /* | ||
196 | * __fsnotify_update_child_dentry_flags(inode); | ||
197 | * | ||
198 | * I really want to call that, but we can't, we have no idea if the inode | ||
199 | * still exists the second we drop the mark->lock. | ||
200 | * | ||
201 | * The next time an event arrive to this inode from one of it's children | ||
202 | * __fsnotify_parent will see that the inode doesn't care about it's | ||
203 | * children and will update all of these flags then. So really this | ||
204 | * is just a lazy update (and could be a perf win...) | ||
205 | */ | ||
206 | |||
207 | |||
208 | iput(inode); | ||
209 | |||
210 | /* | ||
211 | * it's possible that this group tried to destroy itself, but this | ||
212 | * this mark was simultaneously being freed by inode. If that's the | ||
213 | * case, we finish freeing the group here. | ||
214 | */ | ||
215 | if (unlikely(atomic_dec_and_test(&group->num_marks))) | ||
216 | fsnotify_final_destroy_group(group); | ||
217 | } | ||
218 | |||
219 | /* | ||
220 | * Given a group, destroy all of the marks associated with that group. | ||
221 | */ | ||
222 | void fsnotify_clear_marks_by_group(struct fsnotify_group *group) | ||
223 | { | ||
224 | struct fsnotify_mark *lmark, *mark; | ||
225 | LIST_HEAD(free_list); | ||
226 | |||
227 | spin_lock(&group->mark_lock); | ||
228 | list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) { | ||
229 | list_add(&mark->free_g_list, &free_list); | ||
230 | list_del_init(&mark->g_list); | ||
231 | fsnotify_get_mark(mark); | ||
232 | } | ||
233 | spin_unlock(&group->mark_lock); | ||
234 | |||
235 | list_for_each_entry_safe(mark, lmark, &free_list, free_g_list) { | ||
236 | fsnotify_destroy_mark(mark); | ||
237 | fsnotify_put_mark(mark); | ||
238 | } | ||
239 | } | 81 | } |
240 | 82 | ||
241 | /* | 83 | /* |
@@ -261,8 +103,12 @@ void fsnotify_clear_marks_by_inode(struct inode *inode) | |||
261 | } | 103 | } |
262 | } | 104 | } |
263 | 105 | ||
264 | static struct fsnotify_mark *fsnotify_find_mark_locked(struct fsnotify_group *group, | 106 | /* |
265 | struct inode *inode) | 107 | * given a group and inode, find the mark associated with that combination. |
108 | * if found take a reference to that mark and return it, else return NULL | ||
109 | */ | ||
110 | struct fsnotify_mark *fsnotify_find_inode_mark_locked(struct fsnotify_group *group, | ||
111 | struct inode *inode) | ||
266 | { | 112 | { |
267 | struct fsnotify_mark *mark; | 113 | struct fsnotify_mark *mark; |
268 | struct hlist_node *pos; | 114 | struct hlist_node *pos; |
@@ -282,50 +128,26 @@ static struct fsnotify_mark *fsnotify_find_mark_locked(struct fsnotify_group *gr | |||
282 | * given a group and inode, find the mark associated with that combination. | 128 | * given a group and inode, find the mark associated with that combination. |
283 | * if found take a reference to that mark and return it, else return NULL | 129 | * if found take a reference to that mark and return it, else return NULL |
284 | */ | 130 | */ |
285 | struct fsnotify_mark *fsnotify_find_mark(struct fsnotify_group *group, | 131 | struct fsnotify_mark *fsnotify_find_inode_mark(struct fsnotify_group *group, |
286 | struct inode *inode) | 132 | struct inode *inode) |
287 | { | 133 | { |
288 | struct fsnotify_mark *mark; | 134 | struct fsnotify_mark *mark; |
289 | 135 | ||
290 | spin_lock(&inode->i_lock); | 136 | spin_lock(&inode->i_lock); |
291 | mark = fsnotify_find_mark_locked(group, inode); | 137 | mark = fsnotify_find_inode_mark_locked(group, inode); |
292 | spin_unlock(&inode->i_lock); | 138 | spin_unlock(&inode->i_lock); |
293 | 139 | ||
294 | return mark; | 140 | return mark; |
295 | } | 141 | } |
296 | 142 | ||
297 | void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old) | ||
298 | { | ||
299 | assert_spin_locked(&old->lock); | ||
300 | new->i.inode = old->i.inode; | ||
301 | new->group = old->group; | ||
302 | new->mask = old->mask; | ||
303 | new->free_mark = old->free_mark; | ||
304 | } | ||
305 | |||
306 | /* | ||
307 | * Nothing fancy, just initialize lists and locks and counters. | ||
308 | */ | ||
309 | void fsnotify_init_mark(struct fsnotify_mark *mark, | ||
310 | void (*free_mark)(struct fsnotify_mark *mark)) | ||
311 | { | ||
312 | spin_lock_init(&mark->lock); | ||
313 | atomic_set(&mark->refcnt, 1); | ||
314 | INIT_HLIST_NODE(&mark->i.i_list); | ||
315 | mark->group = NULL; | ||
316 | mark->mask = 0; | ||
317 | mark->i.inode = NULL; | ||
318 | mark->free_mark = free_mark; | ||
319 | } | ||
320 | |||
321 | /* | 143 | /* |
322 | * Attach an initialized mark mark to a given group and inode. | 144 | * Attach an initialized mark mark to a given group and inode. |
323 | * These marks may be used for the fsnotify backend to determine which | 145 | * These marks may be used for the fsnotify backend to determine which |
324 | * event types should be delivered to which group and for which inodes. | 146 | * event types should be delivered to which group and for which inodes. |
325 | */ | 147 | */ |
326 | int fsnotify_add_mark(struct fsnotify_mark *mark, | 148 | int fsnotify_add_inode_mark(struct fsnotify_mark *mark, |
327 | struct fsnotify_group *group, struct inode *inode, | 149 | struct fsnotify_group *group, struct inode *inode, |
328 | int allow_dups) | 150 | int allow_dups) |
329 | { | 151 | { |
330 | struct fsnotify_mark *lmark = NULL; | 152 | struct fsnotify_mark *lmark = NULL; |
331 | int ret = 0; | 153 | int ret = 0; |
@@ -336,56 +158,26 @@ int fsnotify_add_mark(struct fsnotify_mark *mark, | |||
336 | 158 | ||
337 | mark->flags = FSNOTIFY_MARK_FLAG_INODE; | 159 | mark->flags = FSNOTIFY_MARK_FLAG_INODE; |
338 | 160 | ||
339 | /* | 161 | assert_spin_locked(&mark->lock); |
340 | * if this group isn't being testing for inode type events we need | 162 | assert_spin_locked(&group->mark_lock); |
341 | * to start testing | ||
342 | */ | ||
343 | if (unlikely(list_empty(&group->inode_group_list))) | ||
344 | fsnotify_add_inode_group(group); | ||
345 | /* | ||
346 | * XXX This is where we could also do the fsnotify_add_vfsmount_group | ||
347 | * if we are setting and vfsmount mark.... | ||
348 | |||
349 | if (unlikely(list_empty(&group->vfsmount_group_list))) | ||
350 | fsnotify_add_vfsmount_group(group); | ||
351 | */ | ||
352 | 163 | ||
353 | /* | ||
354 | * LOCKING ORDER!!!! | ||
355 | * mark->lock | ||
356 | * group->mark_lock | ||
357 | * inode->i_lock | ||
358 | */ | ||
359 | spin_lock(&mark->lock); | ||
360 | spin_lock(&group->mark_lock); | ||
361 | spin_lock(&inode->i_lock); | 164 | spin_lock(&inode->i_lock); |
362 | 165 | ||
363 | if (!allow_dups) | 166 | if (!allow_dups) |
364 | lmark = fsnotify_find_mark_locked(group, inode); | 167 | lmark = fsnotify_find_inode_mark_locked(group, inode); |
365 | if (!lmark) { | 168 | if (!lmark) { |
366 | mark->group = group; | ||
367 | mark->i.inode = inode; | 169 | mark->i.inode = inode; |
368 | 170 | ||
369 | hlist_add_head(&mark->i.i_list, &inode->i_fsnotify_marks); | 171 | hlist_add_head(&mark->i.i_list, &inode->i_fsnotify_marks); |
370 | list_add(&mark->g_list, &group->marks_list); | ||
371 | |||
372 | fsnotify_get_mark(mark); /* for i_list and g_list */ | ||
373 | |||
374 | atomic_inc(&group->num_marks); | ||
375 | 172 | ||
376 | fsnotify_recalc_inode_mask_locked(inode); | 173 | fsnotify_recalc_inode_mask_locked(inode); |
377 | } | 174 | } |
378 | 175 | ||
379 | spin_unlock(&inode->i_lock); | 176 | spin_unlock(&inode->i_lock); |
380 | spin_unlock(&group->mark_lock); | ||
381 | spin_unlock(&mark->lock); | ||
382 | 177 | ||
383 | if (lmark) { | 178 | if (lmark) { |
384 | ret = -EEXIST; | 179 | ret = -EEXIST; |
385 | iput(inode); | 180 | iput(inode); |
386 | fsnotify_put_mark(lmark); | ||
387 | } else { | ||
388 | __fsnotify_update_child_dentry_flags(inode); | ||
389 | } | 181 | } |
390 | 182 | ||
391 | return ret; | 183 | return ret; |