* [PATCH v4 0/2] erofs: support large folios for fscache mode @ 2022-12-01 7:42 Jingbo Xu 2022-12-01 7:42 ` [PATCH v4 1/2] " Jingbo Xu 2022-12-01 7:42 ` [PATCH v4 2/2] erofs: enable " Jingbo Xu 0 siblings, 2 replies; 4+ messages in thread From: Jingbo Xu @ 2022-12-01 7:42 UTC (permalink / raw) To: xiang, chao, linux-erofs; +Cc: linux-kernel v4: - patch 1: move the change to the INLINE routine to a separate patch which will be sent laterly - patch 1: introduce erofs_fscache_req_chain() helper to improve the code organization - patch 1: misc improvement to make the code cleaner - patch 2: also document the new feature in documentation v3: https://lore.kernel.org/all/20221129115833.41062-1-jefflexu@linux.alibaba.com/ v2: https://lore.kernel.org/all/20221128025011.36352-1-jefflexu@linux.alibaba.com/ v1: https://lore.kernel.org/all/20221126005756.7662-1-jefflexu@linux.alibaba.com/ v3: - patch 1: when large folios supported, one folio or folio range can be mapped into several slices, with each slice mapped to different cookies, and thus each slice needs its own netfs_cache_resources. In the implementation of v2, each .read_folio() or .readahead() calling corresponds to only one request, and thus only one netfs_cache_resources (embedded in the request). In this case, fscache_begin_read_operation() may be called multiple times on this cres, while cres->ops->end_operation() is called only once when the whole request completes. This can cause the leakage of the corresponding cachefiles_object->n_accesses refcount, which will cause the user daemon hangs there forever waiting for cache->object_count decreasing to 0 when the user daemon exits. Worsely, as we mentioned previously, when large folios supported, one folio or folio range can be mapped to multiple chunks on different cookies, in which case each mapped chunk needs its own cres. In the implementation of v2, each .read_folio() or .readahead() calling corresponds to only one request, and thus only one cres. This will make the only cres used by the first chunk gets overridden by the following chunk. To fix this, we introduce listed requests, where each .read_folio() or .readahead() calling can correspond to a list of requests, with each request corresponds to one cres. v2: - patch 2: keep the enabling for iomap and fscache mode in separate patches; don't enable the feature for the meta data routine for now (Gao Xiang) Patch 1 is the main part of supporting large folios for fscache mode. It relies on a pending patch[1] adding .prepare_ondemand_read() interface in Cachefiles. Patch 2 just turns the switch on and enables the feature for fscache mode. It relies on a previous patch[2] which enables this feature for iomap mode. [1] https://lore.kernel.org/all/20221124034212.81892-1-jefflexu@linux.alibaba.com/ [2] https://lore.kernel.org/all/20221110074023.8059-1-jefflexu@linux.alibaba.com/ Jingbo Xu (2): erofs: support large folios for fscache mode erofs: enable large folios for fscache mode Documentation/filesystems/erofs.rst | 2 + fs/erofs/fscache.c | 148 +++++++++++++++------------- fs/erofs/inode.c | 3 +- 3 files changed, 83 insertions(+), 70 deletions(-) -- 2.19.1.6.gb485710b ^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH v4 1/2] erofs: support large folios for fscache mode 2022-12-01 7:42 [PATCH v4 0/2] erofs: support large folios for fscache mode Jingbo Xu @ 2022-12-01 7:42 ` Jingbo Xu 2022-12-01 9:55 ` [Phishing Risk] [External] " Jia Zhu 2022-12-01 7:42 ` [PATCH v4 2/2] erofs: enable " Jingbo Xu 1 sibling, 1 reply; 4+ messages in thread From: Jingbo Xu @ 2022-12-01 7:42 UTC (permalink / raw) To: xiang, chao, linux-erofs; +Cc: linux-kernel When large folios supported, one folio can be split into several slices, each of which may be mapped to META/UNMAPPED/MAPPED, and the folio can be unlocked as a whole only when all slices have completed. Thus always allocate erofs_fscache_request for each .read_folio() or .readahead(), in which case the allocated request is responsible for unlocking folios when all slices have completed. As described above, each folio or folio range can be mapped into several slices, while these slices may be mapped to different cookies, and thus each slice needs its own netfs_cache_resources. Here we introduce chained requests to support this, where each .read_folio() or .readahead() calling can correspond to multiple requests. Each request has its own netfs_cache_resources and thus is used to access one cookie. Among these requests, there's a primary request, with the others pointing to the primary request. Signed-off-by: Jingbo Xu <jefflexu@linux.alibaba.com> --- fs/erofs/fscache.c | 148 ++++++++++++++++++++++++--------------------- 1 file changed, 80 insertions(+), 68 deletions(-) diff --git a/fs/erofs/fscache.c b/fs/erofs/fscache.c index 3e794891cd91..f14886c479bd 100644 --- a/fs/erofs/fscache.c +++ b/fs/erofs/fscache.c @@ -12,6 +12,7 @@ static LIST_HEAD(erofs_domain_list); static struct vfsmount *erofs_pseudo_mnt; struct erofs_fscache_request { + struct erofs_fscache_request *primary; struct netfs_cache_resources cache_resources; struct address_space *mapping; /* The mapping being accessed */ loff_t start; /* Start position */ @@ -38,6 +39,26 @@ static struct erofs_fscache_request *erofs_fscache_req_alloc(struct address_spac return req; } +static struct erofs_fscache_request *erofs_fscache_req_chain(struct erofs_fscache_request *primary, + size_t len) +{ + struct erofs_fscache_request *req; + + /* use primary request for the first submission */ + if (!primary->submitted) { + refcount_inc(&primary->ref); + return primary; + } + + req = erofs_fscache_req_alloc(primary->mapping, + primary->start + primary->submitted, len); + if (!IS_ERR(req)) { + req->primary = primary; + refcount_inc(&primary->ref); + } + return req; +} + static void erofs_fscache_req_complete(struct erofs_fscache_request *req) { struct folio *folio; @@ -56,17 +77,19 @@ static void erofs_fscache_req_complete(struct erofs_fscache_request *req) folio_unlock(folio); } rcu_read_unlock(); - - if (req->cache_resources.ops) - req->cache_resources.ops->end_operation(&req->cache_resources); - - kfree(req); } static void erofs_fscache_req_put(struct erofs_fscache_request *req) { - if (refcount_dec_and_test(&req->ref)) - erofs_fscache_req_complete(req); + if (refcount_dec_and_test(&req->ref)) { + if (req->cache_resources.ops) + req->cache_resources.ops->end_operation(&req->cache_resources); + if (!req->primary) + erofs_fscache_req_complete(req); + else + erofs_fscache_req_put(req->primary); + kfree(req); + } } static void erofs_fscache_subreq_complete(void *priv, @@ -74,8 +97,12 @@ static void erofs_fscache_subreq_complete(void *priv, { struct erofs_fscache_request *req = priv; - if (IS_ERR_VALUE(transferred_or_error)) - req->error = transferred_or_error; + if (IS_ERR_VALUE(transferred_or_error)) { + if (req->primary) + req->primary->error = transferred_or_error; + else + req->error = transferred_or_error; + } erofs_fscache_req_put(req); } @@ -131,7 +158,6 @@ static int erofs_fscache_read_folios_async(struct fscache_cookie *cookie, done += slen; } DBG_BUGON(done != len); - req->submitted += len; return 0; } @@ -167,32 +193,19 @@ static int erofs_fscache_meta_read_folio(struct file *data, struct folio *folio) return ret; } -/* - * Read into page cache in the range described by (@pos, @len). - * - * On return, if the output @unlock is true, the caller is responsible for page - * unlocking; otherwise the callee will take this responsibility through request - * completion. - * - * The return value is the number of bytes successfully handled, or negative - * error code on failure. The only exception is that, the length of the range - * instead of the error code is returned on failure after request is allocated, - * so that .readahead() could advance rac accordingly. - */ -static int erofs_fscache_data_read(struct address_space *mapping, - loff_t pos, size_t len, bool *unlock) +static int erofs_fscache_data_read_slice(struct erofs_fscache_request *primary) { + struct address_space *mapping = primary->mapping; struct inode *inode = mapping->host; struct super_block *sb = inode->i_sb; struct erofs_fscache_request *req; struct erofs_map_blocks map; struct erofs_map_dev mdev; struct iov_iter iter; + loff_t pos = primary->start + primary->submitted; size_t count; int ret; - *unlock = true; - map.m_la = pos; ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW); if (ret) @@ -220,17 +233,19 @@ static int erofs_fscache_data_read(struct address_space *mapping, } iov_iter_zero(PAGE_SIZE - size, &iter); erofs_put_metabuf(&buf); - return PAGE_SIZE; + primary->submitted += PAGE_SIZE; + return 0; } + count = primary->len - primary->submitted; if (!(map.m_flags & EROFS_MAP_MAPPED)) { - count = len; iov_iter_xarray(&iter, READ, &mapping->i_pages, pos, count); iov_iter_zero(count, &iter); - return count; + primary->submitted += count; + return 0; } - count = min_t(size_t, map.m_llen - (pos - map.m_la), len); + count = min_t(size_t, map.m_llen - (pos - map.m_la), count); DBG_BUGON(!count || count % PAGE_SIZE); mdev = (struct erofs_map_dev) { @@ -241,68 +256,65 @@ static int erofs_fscache_data_read(struct address_space *mapping, if (ret) return ret; - req = erofs_fscache_req_alloc(mapping, pos, count); + req = erofs_fscache_req_chain(primary, count); if (IS_ERR(req)) return PTR_ERR(req); - *unlock = false; ret = erofs_fscache_read_folios_async(mdev.m_fscache->cookie, req, mdev.m_pa + (pos - map.m_la), count); - if (ret) - req->error = ret; - erofs_fscache_req_put(req); - return count; + primary->submitted += count; + return ret; } -static int erofs_fscache_read_folio(struct file *file, struct folio *folio) +static int erofs_fscache_data_read(struct erofs_fscache_request *req) { - bool unlock; int ret; - DBG_BUGON(folio_size(folio) != EROFS_BLKSIZ); + do { + ret = erofs_fscache_data_read_slice(req); + if (ret) + req->error = ret; + } while (!ret && req->submitted < req->len); - ret = erofs_fscache_data_read(folio_mapping(folio), folio_pos(folio), - folio_size(folio), &unlock); - if (unlock) { - if (ret > 0) - folio_mark_uptodate(folio); + return ret; +} + +static int erofs_fscache_read_folio(struct file *file, struct folio *folio) +{ + struct erofs_fscache_request *req; + int ret; + + req = erofs_fscache_req_alloc(folio_mapping(folio), + folio_pos(folio), folio_size(folio)); + if (IS_ERR(req)) { folio_unlock(folio); + return PTR_ERR(req); } - return ret < 0 ? ret : 0; + + ret = erofs_fscache_data_read(req); + erofs_fscache_req_put(req); + return ret; } static void erofs_fscache_readahead(struct readahead_control *rac) { - struct folio *folio; - size_t len, done = 0; - loff_t start, pos; - bool unlock; - int ret, size; + struct erofs_fscache_request *req; if (!readahead_count(rac)) return; - start = readahead_pos(rac); - len = readahead_length(rac); + req = erofs_fscache_req_alloc(rac->mapping, + readahead_pos(rac), readahead_length(rac)); + if (IS_ERR(req)) + return; - do { - pos = start + done; - ret = erofs_fscache_data_read(rac->mapping, pos, - len - done, &unlock); - if (ret <= 0) - return; + /* The request completion will drop refs on the folios. */ + while (readahead_folio(rac)) + ; - size = ret; - while (size) { - folio = readahead_folio(rac); - size -= folio_size(folio); - if (unlock) { - folio_mark_uptodate(folio); - folio_unlock(folio); - } - } - } while ((done += ret) < len); + erofs_fscache_data_read(req); + erofs_fscache_req_put(req); } static const struct address_space_operations erofs_fscache_meta_aops = { -- 2.19.1.6.gb485710b ^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [Phishing Risk] [External] [PATCH v4 1/2] erofs: support large folios for fscache mode 2022-12-01 7:42 ` [PATCH v4 1/2] " Jingbo Xu @ 2022-12-01 9:55 ` Jia Zhu 0 siblings, 0 replies; 4+ messages in thread From: Jia Zhu @ 2022-12-01 9:55 UTC (permalink / raw) To: Jingbo Xu, xiang, chao, linux-erofs; +Cc: linux-kernel 在 2022/12/1 15:42, Jingbo Xu 写道: > When large folios supported, one folio can be split into several slices, > each of which may be mapped to META/UNMAPPED/MAPPED, and the folio can > be unlocked as a whole only when all slices have completed. > > Thus always allocate erofs_fscache_request for each .read_folio() or > .readahead(), in which case the allocated request is responsible for > unlocking folios when all slices have completed. > > As described above, each folio or folio range can be mapped into several > slices, while these slices may be mapped to different cookies, and thus > each slice needs its own netfs_cache_resources. Here we introduce > chained requests to support this, where each .read_folio() or > .readahead() calling can correspond to multiple requests. Each request > has its own netfs_cache_resources and thus is used to access one cookie. > Among these requests, there's a primary request, with the others > pointing to the primary request. > > Signed-off-by: Jingbo Xu <jefflexu@linux.alibaba.com> Reviewed-by: Jia Zhu <zhujia.zj@bytedance.com> Thanks. > --- > fs/erofs/fscache.c | 148 ++++++++++++++++++++++++--------------------- > 1 file changed, 80 insertions(+), 68 deletions(-) > > diff --git a/fs/erofs/fscache.c b/fs/erofs/fscache.c > index 3e794891cd91..f14886c479bd 100644 > --- a/fs/erofs/fscache.c > +++ b/fs/erofs/fscache.c > @@ -12,6 +12,7 @@ static LIST_HEAD(erofs_domain_list); > static struct vfsmount *erofs_pseudo_mnt; > > struct erofs_fscache_request { > + struct erofs_fscache_request *primary; > struct netfs_cache_resources cache_resources; > struct address_space *mapping; /* The mapping being accessed */ > loff_t start; /* Start position */ > @@ -38,6 +39,26 @@ static struct erofs_fscache_request *erofs_fscache_req_alloc(struct address_spac > return req; > } > > +static struct erofs_fscache_request *erofs_fscache_req_chain(struct erofs_fscache_request *primary, > + size_t len) > +{ > + struct erofs_fscache_request *req; > + > + /* use primary request for the first submission */ > + if (!primary->submitted) { > + refcount_inc(&primary->ref); > + return primary; > + } > + > + req = erofs_fscache_req_alloc(primary->mapping, > + primary->start + primary->submitted, len); > + if (!IS_ERR(req)) { > + req->primary = primary; > + refcount_inc(&primary->ref); > + } > + return req; > +} > + > static void erofs_fscache_req_complete(struct erofs_fscache_request *req) > { > struct folio *folio; > @@ -56,17 +77,19 @@ static void erofs_fscache_req_complete(struct erofs_fscache_request *req) > folio_unlock(folio); > } > rcu_read_unlock(); > - > - if (req->cache_resources.ops) > - req->cache_resources.ops->end_operation(&req->cache_resources); > - > - kfree(req); > } > > static void erofs_fscache_req_put(struct erofs_fscache_request *req) > { > - if (refcount_dec_and_test(&req->ref)) > - erofs_fscache_req_complete(req); > + if (refcount_dec_and_test(&req->ref)) { > + if (req->cache_resources.ops) > + req->cache_resources.ops->end_operation(&req->cache_resources); > + if (!req->primary) > + erofs_fscache_req_complete(req); > + else > + erofs_fscache_req_put(req->primary); > + kfree(req); > + } > } > > static void erofs_fscache_subreq_complete(void *priv, > @@ -74,8 +97,12 @@ static void erofs_fscache_subreq_complete(void *priv, > { > struct erofs_fscache_request *req = priv; > > - if (IS_ERR_VALUE(transferred_or_error)) > - req->error = transferred_or_error; > + if (IS_ERR_VALUE(transferred_or_error)) { > + if (req->primary) > + req->primary->error = transferred_or_error; > + else > + req->error = transferred_or_error; > + } > erofs_fscache_req_put(req); > } > > @@ -131,7 +158,6 @@ static int erofs_fscache_read_folios_async(struct fscache_cookie *cookie, > done += slen; > } > DBG_BUGON(done != len); > - req->submitted += len; > return 0; > } > > @@ -167,32 +193,19 @@ static int erofs_fscache_meta_read_folio(struct file *data, struct folio *folio) > return ret; > } > > -/* > - * Read into page cache in the range described by (@pos, @len). > - * > - * On return, if the output @unlock is true, the caller is responsible for page > - * unlocking; otherwise the callee will take this responsibility through request > - * completion. > - * > - * The return value is the number of bytes successfully handled, or negative > - * error code on failure. The only exception is that, the length of the range > - * instead of the error code is returned on failure after request is allocated, > - * so that .readahead() could advance rac accordingly. > - */ > -static int erofs_fscache_data_read(struct address_space *mapping, > - loff_t pos, size_t len, bool *unlock) > +static int erofs_fscache_data_read_slice(struct erofs_fscache_request *primary) > { > + struct address_space *mapping = primary->mapping; > struct inode *inode = mapping->host; > struct super_block *sb = inode->i_sb; > struct erofs_fscache_request *req; > struct erofs_map_blocks map; > struct erofs_map_dev mdev; > struct iov_iter iter; > + loff_t pos = primary->start + primary->submitted; > size_t count; > int ret; > > - *unlock = true; > - > map.m_la = pos; > ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW); > if (ret) > @@ -220,17 +233,19 @@ static int erofs_fscache_data_read(struct address_space *mapping, > } > iov_iter_zero(PAGE_SIZE - size, &iter); > erofs_put_metabuf(&buf); > - return PAGE_SIZE; > + primary->submitted += PAGE_SIZE; > + return 0; > } > > + count = primary->len - primary->submitted; > if (!(map.m_flags & EROFS_MAP_MAPPED)) { > - count = len; > iov_iter_xarray(&iter, READ, &mapping->i_pages, pos, count); > iov_iter_zero(count, &iter); > - return count; > + primary->submitted += count; > + return 0; > } > > - count = min_t(size_t, map.m_llen - (pos - map.m_la), len); > + count = min_t(size_t, map.m_llen - (pos - map.m_la), count); > DBG_BUGON(!count || count % PAGE_SIZE); > > mdev = (struct erofs_map_dev) { > @@ -241,68 +256,65 @@ static int erofs_fscache_data_read(struct address_space *mapping, > if (ret) > return ret; > > - req = erofs_fscache_req_alloc(mapping, pos, count); > + req = erofs_fscache_req_chain(primary, count); > if (IS_ERR(req)) > return PTR_ERR(req); > > - *unlock = false; > ret = erofs_fscache_read_folios_async(mdev.m_fscache->cookie, > req, mdev.m_pa + (pos - map.m_la), count); > - if (ret) > - req->error = ret; > - > erofs_fscache_req_put(req); > - return count; > + primary->submitted += count; > + return ret; > } > > -static int erofs_fscache_read_folio(struct file *file, struct folio *folio) > +static int erofs_fscache_data_read(struct erofs_fscache_request *req) > { > - bool unlock; > int ret; > > - DBG_BUGON(folio_size(folio) != EROFS_BLKSIZ); > + do { > + ret = erofs_fscache_data_read_slice(req); > + if (ret) > + req->error = ret; > + } while (!ret && req->submitted < req->len); > > - ret = erofs_fscache_data_read(folio_mapping(folio), folio_pos(folio), > - folio_size(folio), &unlock); > - if (unlock) { > - if (ret > 0) > - folio_mark_uptodate(folio); > + return ret; > +} > + > +static int erofs_fscache_read_folio(struct file *file, struct folio *folio) > +{ > + struct erofs_fscache_request *req; > + int ret; > + > + req = erofs_fscache_req_alloc(folio_mapping(folio), > + folio_pos(folio), folio_size(folio)); > + if (IS_ERR(req)) { > folio_unlock(folio); > + return PTR_ERR(req); > } > - return ret < 0 ? ret : 0; > + > + ret = erofs_fscache_data_read(req); > + erofs_fscache_req_put(req); > + return ret; > } > > static void erofs_fscache_readahead(struct readahead_control *rac) > { > - struct folio *folio; > - size_t len, done = 0; > - loff_t start, pos; > - bool unlock; > - int ret, size; > + struct erofs_fscache_request *req; > > if (!readahead_count(rac)) > return; > > - start = readahead_pos(rac); > - len = readahead_length(rac); > + req = erofs_fscache_req_alloc(rac->mapping, > + readahead_pos(rac), readahead_length(rac)); > + if (IS_ERR(req)) > + return; > > - do { > - pos = start + done; > - ret = erofs_fscache_data_read(rac->mapping, pos, > - len - done, &unlock); > - if (ret <= 0) > - return; > + /* The request completion will drop refs on the folios. */ > + while (readahead_folio(rac)) > + ; > > - size = ret; > - while (size) { > - folio = readahead_folio(rac); > - size -= folio_size(folio); > - if (unlock) { > - folio_mark_uptodate(folio); > - folio_unlock(folio); > - } > - } > - } while ((done += ret) < len); > + erofs_fscache_data_read(req); > + erofs_fscache_req_put(req); > } > > static const struct address_space_operations erofs_fscache_meta_aops = { ^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH v4 2/2] erofs: enable large folios for fscache mode 2022-12-01 7:42 [PATCH v4 0/2] erofs: support large folios for fscache mode Jingbo Xu 2022-12-01 7:42 ` [PATCH v4 1/2] " Jingbo Xu @ 2022-12-01 7:42 ` Jingbo Xu 1 sibling, 0 replies; 4+ messages in thread From: Jingbo Xu @ 2022-12-01 7:42 UTC (permalink / raw) To: xiang, chao, linux-erofs; +Cc: linux-kernel Enable large folios for fscache mode. Enable this feature for non-compressed format for now, until the compression part supports large folios later. One thing worth noting is that, the feature is not enabled for the meta data routine since meta inodes don't need large folios for now, nor do they support readahead yet. Also document this new feature. Signed-off-by: Jingbo Xu <jefflexu@linux.alibaba.com> Reviewed-by: Jia Zhu <zhujia.zj@bytedance.com> --- Documentation/filesystems/erofs.rst | 2 ++ fs/erofs/inode.c | 3 +-- 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/Documentation/filesystems/erofs.rst b/Documentation/filesystems/erofs.rst index 82af67fdaf99..1c1f7404b338 100644 --- a/Documentation/filesystems/erofs.rst +++ b/Documentation/filesystems/erofs.rst @@ -72,6 +72,8 @@ Here are the main features of EROFS: - Support merging tail-end data into a special inode as fragments. + - Support large folios for uncompressed files. + - Support direct I/O on uncompressed files to avoid double caching for loop devices; diff --git a/fs/erofs/inode.c b/fs/erofs/inode.c index e457b8a59ee7..85932086d23f 100644 --- a/fs/erofs/inode.c +++ b/fs/erofs/inode.c @@ -295,8 +295,7 @@ static int erofs_fill_inode(struct inode *inode) goto out_unlock; } inode->i_mapping->a_ops = &erofs_raw_access_aops; - if (!erofs_is_fscache_mode(inode->i_sb)) - mapping_set_large_folios(inode->i_mapping); + mapping_set_large_folios(inode->i_mapping); #ifdef CONFIG_EROFS_FS_ONDEMAND if (erofs_is_fscache_mode(inode->i_sb)) inode->i_mapping->a_ops = &erofs_fscache_access_aops; -- 2.19.1.6.gb485710b ^ permalink raw reply related [flat|nested] 4+ messages in thread
end of thread, other threads:[~2022-12-01 9:56 UTC | newest] Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2022-12-01 7:42 [PATCH v4 0/2] erofs: support large folios for fscache mode Jingbo Xu 2022-12-01 7:42 ` [PATCH v4 1/2] " Jingbo Xu 2022-12-01 9:55 ` [Phishing Risk] [External] " Jia Zhu 2022-12-01 7:42 ` [PATCH v4 2/2] erofs: enable " Jingbo Xu
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).