Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752577Ab3JLJOP (ORCPT ); Sat, 12 Oct 2013 05:14:15 -0400 Received: from mail-ie0-f172.google.com ([209.85.223.172]:50368 "EHLO mail-ie0-f172.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750830Ab3JLJOM (ORCPT ); Sat, 12 Oct 2013 05:14:12 -0400 MIME-Version: 1.0 In-Reply-To: References: <000201ceb836$4c549740$e4fdc5c0$%yang@samsung.com> <20130924010308.GG17725@bbox> <000001ceba6a$997d0490$cc770db0$%yang@samsung.com> <20131011071259.GC6847@bbox> Date: Sat, 12 Oct 2013 17:14:12 +0800 Message-ID: Subject: Re: [PATCH v3 2/3] mm/zswap: bugfix: memory leak when invalidate and reclaim occur concurrently From: Weijie Yang To: Bob Liu Cc: Minchan Kim , Weijie Yang , Andrew Morton , Seth Jennings , Bob Liu , Linux-MM , Linux-Kernel , stable@vger.kernel.org, d.j.shin@samsung.com, heesub.shin@samsung.com, Kyungmin Park , hau.chen@samsung.com, bifeng.tong@samsung.com, rui.xie@samsung.com Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 9620 Lines: 257 On Sat, Oct 12, 2013 at 10:50 AM, Bob Liu wrote: > On Fri, Oct 11, 2013 at 3:13 PM, Minchan Kim wrote: >> On Thu, Sep 26, 2013 at 11:42:17AM +0800, Weijie Yang wrote: >>> On Tue, Sep 24, 2013 at 9:03 AM, Minchan Kim wrote: >>> > On Mon, Sep 23, 2013 at 04:21:49PM +0800, Weijie Yang wrote: >>> > > >>> > > Modify: >>> > > - check the refcount in fail path, free memory if it is not referenced. >>> > >>> > Hmm, I don't like this because zswap refcount routine is already mess for me. >>> > I'm not sure why it was designed from the beginning. I hope we should fix it first. >>> > >>> > 1. zswap_rb_serach could include zswap_entry_get semantic if it founds a entry from >>> > the tree. Of course, we should ranme it as find_get_zswap_entry like find_get_page. >>> > 2. zswap_entry_put could hide resource free function like zswap_free_entry so that >>> > all of caller can use it easily following pattern. >>> > >>> > find_get_zswap_entry >>> > ... >>> > ... >>> > zswap_entry_put >>> > >>> > Of course, zswap_entry_put have to check the entry is in the tree or not >>> > so if someone already removes it from the tree, it should avoid double remove. >>> > >>> > One of the concern I can think is that approach extends critical section >>> > but I think it would be no problem because more bottleneck would be [de]compress >>> > functions. If it were really problem, we can mitigate a problem with moving >>> > unnecessary functions out of zswap_free_entry because it seem to be rather >>> > over-enginnering. >>> >>> I refactor the zswap refcount routine according to Minchan's idea. >>> Here is the new patch, Any suggestion is welcomed. >>> >>> To Seth and Bob, would you please review it again? >> >> Yeah, Seth, Bob. You guys are right persons to review this because this >> scheme was suggested by me who is biased so it couldn't be a fair. ;-) >> But anyway, I will review code itself. >> >>> >>> mm/zswap.c | 116 >>> ++++++++++++++++++++++++++++++++++++++++++++++++++++---------------------------------------------------------------- >>> 1 file changed, 52 insertions(+), 64 deletions(-) >>> >>> diff --git a/mm/zswap.c b/mm/zswap.c >>> old mode 100644 >>> new mode 100755 >>> index deda2b6..bd04910 >>> --- a/mm/zswap.c >>> +++ b/mm/zswap.c >>> @@ -217,6 +217,7 @@ static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp) >>> if (!entry) >>> return NULL; >>> entry->refcount = 1; >>> + RB_CLEAR_NODE(&entry->rbnode); >>> return entry; >>> } >>> >>> @@ -232,10 +233,20 @@ static void zswap_entry_get(struct zswap_entry *entry) >>> } >>> >>> /* caller must hold the tree lock */ >>> -static int zswap_entry_put(struct zswap_entry *entry) >>> +static int zswap_entry_put(struct zswap_tree *tree, struct zswap_entry *entry) >> >> Why should we have return value? If we really need it, it mitigates >> get/put semantic's whole point so I'd like to just return void. >> >> Let me see. >> >>> { >>> - entry->refcount--; >>> - return entry->refcount; >>> + int refcount = --entry->refcount; >>> + >>> + if (refcount <= 0) { >> >> Hmm, I don't like minus refcount, really. >> I hope we could do following as >> >> BUG_ON(refcount < 0); >> if (refcount == 0) { >> ... >> } >> >> >> >>> + if (!RB_EMPTY_NODE(&entry->rbnode)) { >>> + rb_erase(&entry->rbnode, &tree->rbroot); >>> + RB_CLEAR_NODE(&entry->rbnode); >> >> Minor, >> You could make new function zswap_rb_del or zswap_rb_remove which detach the node >> from rb tree and clear node because we have already zswap_rb_insert. >> >> >>> + } >>> + >>> + zswap_free_entry(tree, entry); >>> + } >>> + >>> + return refcount; >>> } >>> >>> /********************************* >>> @@ -258,6 +269,17 @@ static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset) >>> return NULL; >>> } >>> >> >> Add function description. >> >>> +static struct zswap_entry *zswap_entry_find_get(struct rb_root *root, pgoff_t offset) >>> +{ >>> + struct zswap_entry *entry = NULL; >>> + >>> + entry = zswap_rb_search(root, offset); >>> + if (entry) >>> + zswap_entry_get(entry); >>> + >>> + return entry; >>> +} >>> + >>> /* >>> * In the case that a entry with the same offset is found, a pointer to >>> * the existing entry is stored in dupentry and the function returns -EEXIST >>> @@ -387,7 +409,7 @@ static void zswap_free_entry(struct zswap_tree *tree, struct zswap_entry *entry) >>> enum zswap_get_swap_ret { >>> ZSWAP_SWAPCACHE_NEW, >>> ZSWAP_SWAPCACHE_EXIST, >>> - ZSWAP_SWAPCACHE_NOMEM >>> + ZSWAP_SWAPCACHE_FAIL, >>> }; >>> >>> /* >>> @@ -401,9 +423,9 @@ enum zswap_get_swap_ret { >>> * added to the swap cache, and returned in retpage. >>> * >>> * If success, the swap cache page is returned in retpage >>> - * Returns 0 if page was already in the swap cache, page is not locked >>> - * Returns 1 if the new page needs to be populated, page is locked >>> - * Returns <0 on error >>> + * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache >>> + * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated, page is locked >>> + * Returns ZSWAP_SWAPCACHE_FAIL on error >>> */ >>> static int zswap_get_swap_cache_page(swp_entry_t entry, >>> struct page **retpage) >>> @@ -475,7 +497,7 @@ static int zswap_get_swap_cache_page(swp_entry_t entry, >>> if (new_page) >>> page_cache_release(new_page); >>> if (!found_page) >>> - return ZSWAP_SWAPCACHE_NOMEM; >>> + return ZSWAP_SWAPCACHE_FAIL; >>> *retpage = found_page; >>> return ZSWAP_SWAPCACHE_EXIST; >>> } >>> @@ -517,23 +539,22 @@ static int zswap_writeback_entry(struct zbud_pool *pool, unsigned long handle) >>> >>> /* find and ref zswap entry */ >>> spin_lock(&tree->lock); >>> - entry = zswap_rb_search(&tree->rbroot, offset); >>> + entry = zswap_entry_find_get(&tree->rbroot, offset); >>> if (!entry) { >>> /* entry was invalidated */ >>> spin_unlock(&tree->lock); >>> return 0; >>> } >>> - zswap_entry_get(entry); >>> spin_unlock(&tree->lock); >>> BUG_ON(offset != entry->offset); >>> >>> /* try to allocate swap cache page */ >>> switch (zswap_get_swap_cache_page(swpentry, &page)) { >>> - case ZSWAP_SWAPCACHE_NOMEM: /* no memory */ >>> + case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */ >>> ret = -ENOMEM; >>> goto fail; >>> >>> - case ZSWAP_SWAPCACHE_EXIST: /* page is unlocked */ >>> + case ZSWAP_SWAPCACHE_EXIST: >> >> Why did you remove comment? >> >>> /* page is already in the swap cache, ignore for now */ >>> page_cache_release(page); >>> ret = -EEXIST; >>> @@ -562,38 +583,28 @@ static int zswap_writeback_entry(struct zbud_pool *pool, unsigned long handle) >>> zswap_written_back_pages++; >>> >>> spin_lock(&tree->lock); >>> - >>> /* drop local reference */ >>> - zswap_entry_put(entry); >>> + refcount = zswap_entry_put(tree, entry); >>> /* drop the initial reference from entry creation */ >>> - refcount = zswap_entry_put(entry); >>> - >>> - /* >>> - * There are three possible values for refcount here: >>> - * (1) refcount is 1, load is in progress, unlink from rbtree, >>> - * load will free >>> - * (2) refcount is 0, (normal case) entry is valid, >>> - * remove from rbtree and free entry >>> - * (3) refcount is -1, invalidate happened during writeback, >>> - * free entry >>> - */ >>> - if (refcount >= 0) { >>> - /* no invalidate yet, remove from rbtree */ >>> + if (refcount > 0) { >>> rb_erase(&entry->rbnode, &tree->rbroot); >>> + RB_CLEAR_NODE(&entry->rbnode); >>> + refcount = zswap_entry_put(tree, entry); >> >> Now, I see why you need return in zswap_entry_put but let's consider again >> because it's really mess to me and it hurts get/put semantic a lot so >> How about this? >> >> spin_lock(&tree->lock); >> /* drop local reference */ >> zswap_entry_put(tree, entry); >> /* >> * In here, we want to free entry but invalidation may free earlier >> * under us so that we should check it again >> */ >> if (entry == zswap_rb_search(&tree->rb_root, offset)) > > Then where is the place unlink entry from rbtree if load was in progress ? zswap_entry_put() have the unlink handle logic > And in the following fail path, return value from zswap_entry_put() is > also used. It is okay even if we return -EAGAIN in the fail path >> /* Yes, it's stable so we should free it */ >> zswap_entry_put(tree, entry); >> >> /* >> * Whether it would be freed by invalidation or writeback, it doesn't >> * matter. Important thing is that it will be freed so there >> * is no point to return -EAGAIN. >> */ >> spin_unlock(&tree->lock); >> return 0; >> > > -- > Regards, > --Bob -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/