proc: tweak comments about 2 stage open and everything
authorAlexey Dobriyan <adobriyan@gmail.com>
Tue, 13 Dec 2016 00:45:22 +0000 (16:45 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 13 Dec 2016 02:55:09 +0000 (18:55 -0800)
Some comments were obsoleted since commit 05c0ae21c034 ("try a saner
locking for pde_opener...").

Some new comments added.

Some confusing comments replaced with equally confusing ones.

Link: http://lkml.kernel.org/r/20161029160231.GD1246@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/proc/inode.c

index 57f548e2eb592055efc8f87f92be0721c5839cca..783bc19644d18d2b5b64de10f10169a6e957cc28 100644 (file)
@@ -138,6 +138,16 @@ static void unuse_pde(struct proc_dir_entry *pde)
 /* pde is locked */
 static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
 {
+       /*
+        * close() (proc_reg_release()) can't delete an entry and proceed:
+        * ->release hook needs to be available at the right moment.
+        *
+        * rmmod (remove_proc_entry() et al) can't delete an entry and proceed:
+        * "struct file" needs to be available at the right moment.
+        *
+        * Therefore, first process to enter this function does ->release() and
+        * signals its completion to the other process which does nothing.
+        */
        if (pdeo->closing) {
                /* somebody else is doing that, just wait */
                DECLARE_COMPLETION_ONSTACK(c);
@@ -152,6 +162,7 @@ static void close_pdeo(struct proc_dir_entry *pde, struct pde_opener *pdeo)
                file = pdeo->file;
                pde->proc_fops->release(file_inode(file), file);
                spin_lock(&pde->pde_unload_lock);
+               /* After ->release. */
                list_del(&pdeo->lh);
                if (pdeo->c)
                        complete(pdeo->c);
@@ -167,6 +178,8 @@ void proc_entry_rundown(struct proc_dir_entry *de)
        if (atomic_add_return(BIAS, &de->in_use) != BIAS)
                wait_for_completion(&c);
 
+       /* ->pde_openers list can't grow from now on. */
+
        spin_lock(&de->pde_unload_lock);
        while (!list_empty(&de->pde_openers)) {
                struct pde_opener *pdeo;
@@ -312,14 +325,15 @@ static int proc_reg_open(struct inode *inode, struct file *file)
        struct pde_opener *pdeo;
 
        /*
-        * What for, you ask? Well, we can have open, rmmod, remove_proc_entry
-        * sequence. ->release won't be called because ->proc_fops will be
-        * cleared. Depending on complexity of ->release, consequences vary.
+        * Ensure that
+        * 1) PDE's ->release hook will be called no matter what
+        *    either normally by close()/->release, or forcefully by
+        *    rmmod/remove_proc_entry.
+        *
+        * 2) rmmod isn't blocked by opening file in /proc and sitting on
+        *    the descriptor (including "rmmod foo </proc/foo" scenario).
         *
-        * We can't wait for mercy when close will be done for real, it's
-        * deadlockable: rmmod foo </proc/foo . So, we're going to do ->release
-        * by hand in remove_proc_entry(). For this, save opener's credentials
-        * for later.
+        * Save every "struct file" with custom ->release hook.
         */
        pdeo = kmalloc(sizeof(struct pde_opener), GFP_KERNEL);
        if (!pdeo)
@@ -340,7 +354,6 @@ static int proc_reg_open(struct inode *inode, struct file *file)
                pdeo->file = file;
                pdeo->closing = false;
                pdeo->c = NULL;
-               /* Strictly for "too late" ->release in proc_reg_release(). */
                spin_lock(&pde->pde_unload_lock);
                list_add(&pdeo->lh, &pde->pde_openers);
                spin_unlock(&pde->pde_unload_lock);