Hi Christoph, David, Since Christoph asked nicely[1] ;-), here are three patches that go on top of the similar patches for bio structs now in the block tree that make the old block direct-IO code use iov_iter_extract_pages() and page pinning. There are three patches: (1) Make page pinning not add or remove a pin to/from the ZERO_PAGE, thereby allowing the dio code to insert zero pages in the middle of dealing with pinned pages. (2) Provide a function to allow an additional pin to be taken on a page we already have pinned (and do nothing for the zero page). (3) Switch direct-io.c over to using page pinning and to use iov_iter_extract_pages() so that pages from non-user-backed iterators aren't pinned. Note that I haven't managed to test this yet as SELinux is refusing to let me mount things like ext2 filesystems on account of it not having xattrs:-/ I've pushed the patches here also: https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/log/?h=iov-old-dio David Link: https://lore.kernel.org/r/ZGxfrOLZ4aN9/MvE@infradead.org/ [1] David Howells (3): mm: Don't pin ZERO_PAGE in pin_user_pages() mm: Provide a function to get an additional pin on a page block: Use iov_iter_extract_pages() and page pinning in direct-io.c fs/direct-io.c | 68 ++++++++++++++++++++++++++++------------------ include/linux/mm.h | 1 + mm/gup.c | 54 +++++++++++++++++++++++++++++++++++- 3 files changed, 95 insertions(+), 28 deletions(-) Make pin_user_pages*() leave the ZERO_PAGE unpinned if it extracts a pointer to it from the page tables and make unpin_user_page*() correspondingly ignore the ZERO_PAGE when unpinning. We don't want to risk overrunning the zero page's refcount as we're only allowed ~2 million pins on it - something that userspace can conceivably trigger. Signed-off-by: David Howells <dhowells@redhat.com> cc: Christoph Hellwig <hch@infradead.org> cc: David Hildenbrand <david@redhat.com> cc: Andrew Morton <akpm@linux-foundation.org> cc: Jens Axboe <axboe@kernel.dk> cc: Al Viro <viro@zeniv.linux.org.uk> cc: Matthew Wilcox <willy@infradead.org> cc: Jan Kara <jack@suse.cz> cc: Jeff Layton <jlayton@kernel.org> cc: Jason Gunthorpe <jgg@nvidia.com> cc: Logan Gunthorpe <logang@deltatee.com> cc: Hillf Danton <hdanton@sina.com> cc: Christian Brauner <brauner@kernel.org> cc: Linus Torvalds <torvalds@linux-foundation.org> cc: linux-fsdevel@vger.kernel.org cc: linux-block@vger.kernel.org cc: linux-kernel@vger.kernel.org cc: linux-mm@kvack.org --- mm/gup.c | 25 ++++++++++++++++++++++++- 1 file changed, 24 insertions(+), 1 deletion(-) diff --git a/mm/gup.c b/mm/gup.c index XXXXXXX..XXXXXXX 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -XXX,XX +XXX,XX @@ static inline void sanity_check_pinned_pages(struct page **pages, struct page *page = *pages; struct folio *folio = page_folio(page); - if (!folio_test_anon(folio)) + if (page == ZERO_PAGE(0) || + !folio_test_anon(folio)) continue; if (!folio_test_large(folio) || folio_test_hugetlb(folio)) VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page), page); @@ -XXX,XX +XXX,XX @@ struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags) else if (flags & FOLL_PIN) { struct folio *folio; + /* + * Don't take a pin on the zero page - it's not going anywhere + * and it is used in a *lot* of places. + */ + if (page == ZERO_PAGE(0)) + return page_folio(ZERO_PAGE(0)); + /* * Can't do FOLL_LONGTERM + FOLL_PIN gup fast path if not in a * right zone, so fail and let the caller fall back to the slow @@ -XXX,XX +XXX,XX @@ struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags) static void gup_put_folio(struct folio *folio, int refs, unsigned int flags) { if (flags & FOLL_PIN) { + if (folio == page_folio(ZERO_PAGE(0))) + return; node_stat_mod_folio(folio, NR_FOLL_PIN_RELEASED, refs); if (folio_test_large(folio)) atomic_sub(refs, &folio->_pincount); @@ -XXX,XX +XXX,XX @@ int __must_check try_grab_page(struct page *page, unsigned int flags) if (flags & FOLL_GET) folio_ref_inc(folio); else if (flags & FOLL_PIN) { + /* + * Don't take a pin on the zero page - it's not going anywhere + * and it is used in a *lot* of places. + */ + if (page == ZERO_PAGE(0)) + return 0; + /* * Similar to try_grab_folio(): be sure to *also* * increment the normal page refcount field at least once, @@ -XXX,XX +XXX,XX @@ EXPORT_SYMBOL_GPL(get_user_pages_fast); * * FOLL_PIN means that the pages must be released via unpin_user_page(). Please * see Documentation/core-api/pin_user_pages.rst for further details. + * + * Note that if the zero_page is amongst the returned pages, it will not have + * pins in it and unpin_user_page() will not remove pins from it. */ int pin_user_pages_fast(unsigned long start, int nr_pages, unsigned int gup_flags, struct page **pages) @@ -XXX,XX +XXX,XX @@ EXPORT_SYMBOL(pin_user_pages); * pin_user_pages_unlocked() is the FOLL_PIN variant of * get_user_pages_unlocked(). Behavior is the same, except that this one sets * FOLL_PIN and rejects FOLL_GET. + * + * Note that if the zero_page is amongst the returned pages, it will not have + * pins in it and unpin_user_page() will not remove pins from it. */ long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, struct page **pages, unsigned int gup_flags) Provide a function to get an additional pin on a page that we already have a pin on. This will be used in fs/direct-io.c when dispatching multiple bios to a page we've extracted from a user-backed iter rather than redoing the extraction. Signed-off-by: David Howells <dhowells@redhat.com> cc: Christoph Hellwig <hch@infradead.org> cc: David Hildenbrand <david@redhat.com> cc: Andrew Morton <akpm@linux-foundation.org> cc: Jens Axboe <axboe@kernel.dk> cc: Al Viro <viro@zeniv.linux.org.uk> cc: Matthew Wilcox <willy@infradead.org> cc: Jan Kara <jack@suse.cz> cc: Jeff Layton <jlayton@kernel.org> cc: Jason Gunthorpe <jgg@nvidia.com> cc: Logan Gunthorpe <logang@deltatee.com> cc: Hillf Danton <hdanton@sina.com> cc: Christian Brauner <brauner@kernel.org> cc: Linus Torvalds <torvalds@linux-foundation.org> cc: linux-fsdevel@vger.kernel.org cc: linux-block@vger.kernel.org cc: linux-kernel@vger.kernel.org cc: linux-mm@kvack.org --- include/linux/mm.h | 1 + mm/gup.c | 29 +++++++++++++++++++++++++++++ 2 files changed, 30 insertions(+) diff --git a/include/linux/mm.h b/include/linux/mm.h index XXXXXXX..XXXXXXX 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -XXX,XX +XXX,XX @@ int get_user_pages_fast(unsigned long start, int nr_pages, unsigned int gup_flags, struct page **pages); int pin_user_pages_fast(unsigned long start, int nr_pages, unsigned int gup_flags, struct page **pages); +void page_get_additional_pin(struct page *page); int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc); int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc, diff --git a/mm/gup.c b/mm/gup.c index XXXXXXX..XXXXXXX 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -XXX,XX +XXX,XX @@ void unpin_user_page(struct page *page) } EXPORT_SYMBOL(unpin_user_page); +/** + * page_get_additional_pin - Try to get an additional pin on a pinned page + * @page: The page to be pinned + * + * Get an additional pin on a page we already have a pin on. Makes no change + * if the page is the zero_page. + */ +void page_get_additional_pin(struct page *page) +{ + struct folio *folio = page_folio(page); + + if (page == ZERO_PAGE(0)) + return; + + /* + * Similar to try_grab_folio(): be sure to *also* increment the normal + * page refcount field at least once, so that the page really is + * pinned. + */ + if (folio_test_large(folio)) { + WARN_ON_ONCE(atomic_read(&folio->_pincount) < 1); + folio_ref_add(folio, 1); + atomic_add(1, &folio->_pincount); + } else { + WARN_ON_ONCE(folio_ref_count(folio) < GUP_PIN_COUNTING_BIAS); + folio_ref_add(folio, GUP_PIN_COUNTING_BIAS); + } +} + static inline struct folio *gup_folio_range_next(struct page *start, unsigned long npages, unsigned long i, unsigned int *ntails) { Change the old block-based direct-I/O code to use iov_iter_extract_pages() to pin user pages or leave kernel pages unpinned rather than taking refs when submitting bios. This makes use of the preceding patches to not take pins on the zero page (thereby allowing insertion of zero pages in with pinned pages) and to get additional pins on pages, allowing an extracted page to be used in multiple bios without having to re-extract it. Signed-off-by: David Howells <dhowells@redhat.com> cc: Christoph Hellwig <hch@infradead.org> cc: David Hildenbrand <david@redhat.com> cc: Andrew Morton <akpm@linux-foundation.org> cc: Jens Axboe <axboe@kernel.dk> cc: Al Viro <viro@zeniv.linux.org.uk> cc: Matthew Wilcox <willy@infradead.org> cc: Jan Kara <jack@suse.cz> cc: Jeff Layton <jlayton@kernel.org> cc: Jason Gunthorpe <jgg@nvidia.com> cc: Logan Gunthorpe <logang@deltatee.com> cc: Hillf Danton <hdanton@sina.com> cc: Christian Brauner <brauner@kernel.org> cc: Linus Torvalds <torvalds@linux-foundation.org> cc: linux-fsdevel@vger.kernel.org cc: linux-block@vger.kernel.org cc: linux-kernel@vger.kernel.org cc: linux-mm@kvack.org --- fs/direct-io.c | 68 ++++++++++++++++++++++++++++++-------------------- 1 file changed, 41 insertions(+), 27 deletions(-) diff --git a/fs/direct-io.c b/fs/direct-io.c index XXXXXXX..XXXXXXX 100644 --- a/fs/direct-io.c +++ b/fs/direct-io.c @@ -XXX,XX +XXX,XX @@ #include "internal.h" /* - * How many user pages to map in one call to get_user_pages(). This determines - * the size of a structure in the slab cache + * How many user pages to map in one call to iov_iter_extract_pages(). This + * determines the size of a structure in the slab cache */ #define DIO_PAGES 64 @@ -XXX,XX +XXX,XX @@ struct dio { struct inode *inode; loff_t i_size; /* i_size when submitted */ dio_iodone_t *end_io; /* IO completion function */ + bool need_unpin; /* T if we need to unpin the pages */ void *private; /* copy from map_bh.b_private */ /* BIO completion state */ spinlock_t bio_lock; /* protects BIO fields below */ - int page_errors; /* errno from get_user_pages() */ + int page_errors; /* err from iov_iter_extract_pages() */ int is_async; /* is IO async ? */ bool defer_completion; /* defer AIO completion to workqueue? */ bool should_dirty; /* if pages should be dirtied */ @@ -XXX,XX +XXX,XX @@ static inline unsigned dio_pages_present(struct dio_submit *sdio) */ static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio) { + struct page **pages = dio->pages; const enum req_op dio_op = dio->opf & REQ_OP_MASK; ssize_t ret; - ret = iov_iter_get_pages2(sdio->iter, dio->pages, LONG_MAX, DIO_PAGES, - &sdio->from); + ret = iov_iter_extract_pages(sdio->iter, &pages, LONG_MAX, + DIO_PAGES, 0, &sdio->from); if (ret < 0 && sdio->blocks_available && dio_op == REQ_OP_WRITE) { - struct page *page = ZERO_PAGE(0); /* * A memory fault, but the filesystem has some outstanding * mapped blocks. We need to use those blocks up to avoid @@ -XXX,XX +XXX,XX @@ static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio) */ if (dio->page_errors == 0) dio->page_errors = ret; - get_page(page); - dio->pages[0] = page; + dio->pages[0] = ZERO_PAGE(0); sdio->head = 0; sdio->tail = 1; sdio->from = 0; @@ -XXX,XX +XXX,XX @@ static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio) /* * Get another userspace page. Returns an ERR_PTR on error. Pages are - * buffered inside the dio so that we can call get_user_pages() against a - * decent number of pages, less frequently. To provide nicer use of the - * L1 cache. + * buffered inside the dio so that we can call iov_iter_extract_pages() + * against a decent number of pages, less frequently. To provide nicer use of + * the L1 cache. */ static inline struct page *dio_get_page(struct dio *dio, struct dio_submit *sdio) @@ -XXX,XX +XXX,XX @@ static inline struct page *dio_get_page(struct dio *dio, return dio->pages[sdio->head]; } +static void dio_pin_page(struct dio *dio, struct page *page) +{ + if (dio->need_unpin) + page_get_additional_pin(page); +} + +static void dio_unpin_page(struct dio *dio, struct page *page) +{ + if (dio->need_unpin) + unpin_user_page(page); +} + /* * dio_complete() - called when all DIO BIO I/O has been completed * @@ -XXX,XX +XXX,XX @@ static inline void dio_bio_submit(struct dio *dio, struct dio_submit *sdio) */ static inline void dio_cleanup(struct dio *dio, struct dio_submit *sdio) { - while (sdio->head < sdio->tail) - put_page(dio->pages[sdio->head++]); + if (dio->need_unpin) + unpin_user_pages(dio->pages + sdio->head, + sdio->tail - sdio->head); } /* @@ -XXX,XX +XXX,XX @@ static inline int dio_new_bio(struct dio *dio, struct dio_submit *sdio, * * Return zero on success. Non-zero means the caller needs to start a new BIO. */ -static inline int dio_bio_add_page(struct dio_submit *sdio) +static inline int dio_bio_add_page(struct dio *dio, struct dio_submit *sdio) { int ret; @@ -XXX,XX +XXX,XX @@ static inline int dio_bio_add_page(struct dio_submit *sdio) */ if ((sdio->cur_page_len + sdio->cur_page_offset) == PAGE_SIZE) sdio->pages_in_io--; - get_page(sdio->cur_page); + dio_pin_page(dio, sdio->cur_page); sdio->final_block_in_bio = sdio->cur_page_block + (sdio->cur_page_len >> sdio->blkbits); ret = 0; @@ -XXX,XX +XXX,XX @@ static inline int dio_send_cur_page(struct dio *dio, struct dio_submit *sdio, goto out; } - if (dio_bio_add_page(sdio) != 0) { + if (dio_bio_add_page(dio, sdio) != 0) { dio_bio_submit(dio, sdio); ret = dio_new_bio(dio, sdio, sdio->cur_page_block, map_bh); if (ret == 0) { - ret = dio_bio_add_page(sdio); + ret = dio_bio_add_page(dio, sdio); BUG_ON(ret != 0); } } @@ -XXX,XX +XXX,XX @@ submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page, */ if (sdio->cur_page) { ret = dio_send_cur_page(dio, sdio, map_bh); - put_page(sdio->cur_page); + dio_unpin_page(dio, sdio->cur_page); sdio->cur_page = NULL; if (ret) return ret; } - get_page(page); /* It is in dio */ + dio_pin_page(dio, page); /* It is in dio */ sdio->cur_page = page; sdio->cur_page_offset = offset; sdio->cur_page_len = len; @@ -XXX,XX +XXX,XX @@ submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page, ret = dio_send_cur_page(dio, sdio, map_bh); if (sdio->bio) dio_bio_submit(dio, sdio); - put_page(sdio->cur_page); + dio_unpin_page(dio, sdio->cur_page); sdio->cur_page = NULL; } return ret; @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, ret = get_more_blocks(dio, sdio, map_bh); if (ret) { - put_page(page); + dio_unpin_page(dio, page); goto out; } if (!buffer_mapped(map_bh)) @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, /* AKPM: eargh, -ENOTBLK is a hack */ if (dio_op == REQ_OP_WRITE) { - put_page(page); + dio_unpin_page(dio, page); return -ENOTBLK; } @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, if (sdio->block_in_file >= i_size_aligned >> blkbits) { /* We hit eof */ - put_page(page); + dio_unpin_page(dio, page); goto out; } zero_user(page, from, 1 << blkbits); @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, sdio->next_block_for_io, map_bh); if (ret) { - put_page(page); + dio_unpin_page(dio, page); goto out; } sdio->next_block_for_io += this_chunk_blocks; @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, break; } - /* Drop the ref which was taken in get_user_pages() */ - put_page(page); + /* Drop the pin which was taken in get_user_pages() */ + dio_unpin_page(dio, page); } out: return ret; @@ -XXX,XX +XXX,XX @@ ssize_t __blockdev_direct_IO(struct kiocb *iocb, struct inode *inode, /* will be released by direct_io_worker */ inode_lock(inode); } + dio->need_unpin = iov_iter_extract_will_pin(iter); /* Once we sampled i_size check for reads beyond EOF */ dio->i_size = i_size_read(inode); @@ -XXX,XX +XXX,XX @@ ssize_t __blockdev_direct_IO(struct kiocb *iocb, struct inode *inode, ret2 = dio_send_cur_page(dio, &sdio, &map_bh); if (retval == 0) retval = ret2; - put_page(sdio.cur_page); + dio_unpin_page(dio, sdio.cur_page); sdio.cur_page = NULL; } if (sdio.bio) Hi Christoph, David, Lorenzo, Here are three patches that go on top of the similar patches for bio structs now in the block tree that make the old block direct-IO code use iov_iter_extract_pages() and page pinning. There are three patches: (1) Make page pinning neither add nor remove a pin to/from a ZERO_PAGE, thereby allowing the dio code to insert zero pages in the middle of dealing with pinned pages. This also mitigates a potential problem whereby userspace could force the overrun the pin counter of a zero page. A pair of functions are provided to wrap the testing of a page or folio to see if it is a zero page. (2) Provide a function to allow an additional pin to be taken on a page we already have pinned (and do nothing for a zero page). (3) Switch direct-io.c over to using page pinning and to use iov_iter_extract_pages() so that pages from non-user-backed iterators aren't pinned. I've pushed the patches here also: https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/log/?h=iov-old-dio David Changes ======= ver #3) - Move is_zero_page() and is_zero_folio() to mm.h for dependency reasons. - Add more comments and adjust the docs about pinning zero pages. - Rename page_get_additional_pin() to folio_add_pin(). - Use is_zero_folio() in folio_add_pin(). - Rename need_unpin to is_pinned in struct dio. ver #2) - Fix use of ZERO_PAGE(). - Add wrappers for testing if a page is a zero page. - Return the zero page obtained, not ZERO_PAGE(0) unconditionally. - Need to set BIO_PAGE_PINNED conditionally, and not BIO_PAGE_REFFED. Link: https://lore.kernel.org/r/ZGxfrOLZ4aN9/MvE@infradead.org/ [1] Link: https://lore.kernel.org/r/20230525155102.87353-1-dhowells@redhat.com/ # v1 Link: https://lore.kernel.org/r/20230525223953.225496-1-dhowells@redhat.com/ # v2 David Howells (3): mm: Don't pin ZERO_PAGE in pin_user_pages() mm: Provide a function to get an additional pin on a page block: Use iov_iter_extract_pages() and page pinning in direct-io.c Documentation/core-api/pin_user_pages.rst | 6 ++ fs/direct-io.c | 72 ++++++++++++++--------- include/linux/mm.h | 27 ++++++++- mm/gup.c | 58 +++++++++++++++++- 4 files changed, 131 insertions(+), 32 deletions(-) Make pin_user_pages*() leave a ZERO_PAGE unpinned if it extracts a pointer to it from the page tables and make unpin_user_page*() correspondingly ignore a ZERO_PAGE when unpinning. We don't want to risk overrunning a zero page's refcount as we're only allowed ~2 million pins on it - something that userspace can conceivably trigger. Add a pair of functions to test whether a page or a folio is a ZERO_PAGE. Signed-off-by: David Howells <dhowells@redhat.com> cc: Christoph Hellwig <hch@infradead.org> cc: David Hildenbrand <david@redhat.com> cc: Lorenzo Stoakes <lstoakes@gmail.com> cc: Andrew Morton <akpm@linux-foundation.org> cc: Jens Axboe <axboe@kernel.dk> cc: Al Viro <viro@zeniv.linux.org.uk> cc: Matthew Wilcox <willy@infradead.org> cc: Jan Kara <jack@suse.cz> cc: Jeff Layton <jlayton@kernel.org> cc: Jason Gunthorpe <jgg@nvidia.com> cc: Logan Gunthorpe <logang@deltatee.com> cc: Hillf Danton <hdanton@sina.com> cc: Christian Brauner <brauner@kernel.org> cc: Linus Torvalds <torvalds@linux-foundation.org> cc: linux-fsdevel@vger.kernel.org cc: linux-block@vger.kernel.org cc: linux-kernel@vger.kernel.org cc: linux-mm@kvack.org --- Notes: ver #3) - Move is_zero_page() and is_zero_folio() to mm.h for dependency reasons. - Add more comments and adjust the docs. ver #2) - Fix use of ZERO_PAGE(). - Add is_zero_page() and is_zero_folio() wrappers. - Return the zero page obtained, not ZERO_PAGE(0) unconditionally. Documentation/core-api/pin_user_pages.rst | 6 +++++ include/linux/mm.h | 26 +++++++++++++++++-- mm/gup.c | 31 ++++++++++++++++++++++- 3 files changed, 60 insertions(+), 3 deletions(-) diff --git a/Documentation/core-api/pin_user_pages.rst b/Documentation/core-api/pin_user_pages.rst index XXXXXXX..XXXXXXX 100644 --- a/Documentation/core-api/pin_user_pages.rst +++ b/Documentation/core-api/pin_user_pages.rst @@ -XXX,XX +XXX,XX @@ pages: This also leads to limitations: there are only 31-10==21 bits available for a counter that increments 10 bits at a time. +* Because of that limitation, special handling is applied to the zero pages + when using FOLL_PIN. We only pretend to pin a zero page - we don't alter its + refcount or pincount at all (it is permanent, so there's no need). The + unpinning functions also don't do anything to a zero page. This is + transparent to the caller. + * Callers must specifically request "dma-pinned tracking of pages". In other words, just calling get_user_pages() will not suffice; a new set of functions, pin_user_page() and related, must be used. diff --git a/include/linux/mm.h b/include/linux/mm.h index XXXXXXX..XXXXXXX 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -XXX,XX +XXX,XX @@ static inline bool page_needs_cow_for_dma(struct vm_area_struct *vma, return page_maybe_dma_pinned(page); } +/** + * is_zero_page - Query if a page is a zero page + * @page: The page to query + * + * This returns true if @page is one of the permanent zero pages. + */ +static inline bool is_zero_page(const struct page *page) +{ + return is_zero_pfn(page_to_pfn(page)); +} + +/** + * is_zero_folio - Query if a folio is a zero page + * @folio: The folio to query + * + * This returns true if @folio is one of the permanent zero pages. + */ +static inline bool is_zero_folio(const struct folio *folio) +{ + return is_zero_page(&folio->page); +} + /* MIGRATE_CMA and ZONE_MOVABLE do not allow pin pages */ #ifdef CONFIG_MIGRATION static inline bool is_longterm_pinnable_page(struct page *page) @@ -XXX,XX +XXX,XX @@ static inline bool is_longterm_pinnable_page(struct page *page) if (mt == MIGRATE_CMA || mt == MIGRATE_ISOLATE) return false; #endif - /* The zero page may always be pinned */ - if (is_zero_pfn(page_to_pfn(page))) + /* The zero page can be "pinned" but gets special handling. */ + if (is_zero_page(page)) return true; /* Coherent device memory must always allow eviction. */ diff --git a/mm/gup.c b/mm/gup.c index XXXXXXX..XXXXXXX 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -XXX,XX +XXX,XX @@ static inline void sanity_check_pinned_pages(struct page **pages, struct page *page = *pages; struct folio *folio = page_folio(page); - if (!folio_test_anon(folio)) + if (is_zero_page(page) || + !folio_test_anon(folio)) continue; if (!folio_test_large(folio) || folio_test_hugetlb(folio)) VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page), page); @@ -XXX,XX +XXX,XX @@ struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags) else if (flags & FOLL_PIN) { struct folio *folio; + /* + * Don't take a pin on the zero page - it's not going anywhere + * and it is used in a *lot* of places. + */ + if (is_zero_page(page)) + return page_folio(page); + /* * Can't do FOLL_LONGTERM + FOLL_PIN gup fast path if not in a * right zone, so fail and let the caller fall back to the slow @@ -XXX,XX +XXX,XX @@ struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags) static void gup_put_folio(struct folio *folio, int refs, unsigned int flags) { if (flags & FOLL_PIN) { + if (is_zero_folio(folio)) + return; node_stat_mod_folio(folio, NR_FOLL_PIN_RELEASED, refs); if (folio_test_large(folio)) atomic_sub(refs, &folio->_pincount); @@ -XXX,XX +XXX,XX @@ int __must_check try_grab_page(struct page *page, unsigned int flags) if (flags & FOLL_GET) folio_ref_inc(folio); else if (flags & FOLL_PIN) { + /* + * Don't take a pin on the zero page - it's not going anywhere + * and it is used in a *lot* of places. + */ + if (is_zero_page(page)) + return 0; + /* * Similar to try_grab_folio(): be sure to *also* * increment the normal page refcount field at least once, @@ -XXX,XX +XXX,XX @@ EXPORT_SYMBOL_GPL(get_user_pages_fast); * * FOLL_PIN means that the pages must be released via unpin_user_page(). Please * see Documentation/core-api/pin_user_pages.rst for further details. + * + * Note that if a zero_page is amongst the returned pages, it will not have + * pins in it and unpin_user_page() will not remove pins from it. */ int pin_user_pages_fast(unsigned long start, int nr_pages, unsigned int gup_flags, struct page **pages) @@ -XXX,XX +XXX,XX @@ EXPORT_SYMBOL_GPL(pin_user_pages_fast); * * FOLL_PIN means that the pages must be released via unpin_user_page(). Please * see Documentation/core-api/pin_user_pages.rst for details. + * + * Note that if a zero_page is amongst the returned pages, it will not have + * pins in it and unpin_user_page*() will not remove pins from it. */ long pin_user_pages_remote(struct mm_struct *mm, unsigned long start, unsigned long nr_pages, @@ -XXX,XX +XXX,XX @@ EXPORT_SYMBOL(pin_user_pages_remote); * * FOLL_PIN means that the pages must be released via unpin_user_page(). Please * see Documentation/core-api/pin_user_pages.rst for details. + * + * Note that if a zero_page is amongst the returned pages, it will not have + * pins in it and unpin_user_page*() will not remove pins from it. */ long pin_user_pages(unsigned long start, unsigned long nr_pages, unsigned int gup_flags, struct page **pages, @@ -XXX,XX +XXX,XX @@ EXPORT_SYMBOL(pin_user_pages); * pin_user_pages_unlocked() is the FOLL_PIN variant of * get_user_pages_unlocked(). Behavior is the same, except that this one sets * FOLL_PIN and rejects FOLL_GET. + * + * Note that if a zero_page is amongst the returned pages, it will not have + * pins in it and unpin_user_page*() will not remove pins from it. */ long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, struct page **pages, unsigned int gup_flags) Provide a function to get an additional pin on a page that we already have a pin on. This will be used in fs/direct-io.c when dispatching multiple bios to a page we've extracted from a user-backed iter rather than redoing the extraction. Signed-off-by: David Howells <dhowells@redhat.com> cc: Christoph Hellwig <hch@infradead.org> cc: David Hildenbrand <david@redhat.com> cc: Lorenzo Stoakes <lstoakes@gmail.com> cc: Andrew Morton <akpm@linux-foundation.org> cc: Jens Axboe <axboe@kernel.dk> cc: Al Viro <viro@zeniv.linux.org.uk> cc: Matthew Wilcox <willy@infradead.org> cc: Jan Kara <jack@suse.cz> cc: Jeff Layton <jlayton@kernel.org> cc: Jason Gunthorpe <jgg@nvidia.com> cc: Logan Gunthorpe <logang@deltatee.com> cc: Hillf Danton <hdanton@sina.com> cc: Christian Brauner <brauner@kernel.org> cc: Linus Torvalds <torvalds@linux-foundation.org> cc: linux-fsdevel@vger.kernel.org cc: linux-block@vger.kernel.org cc: linux-kernel@vger.kernel.org cc: linux-mm@kvack.org --- Notes: ver #3) - Rename to folio_add_pin(). - Change to using is_zero_folio() include/linux/mm.h | 1 + mm/gup.c | 27 +++++++++++++++++++++++++++ 2 files changed, 28 insertions(+) diff --git a/include/linux/mm.h b/include/linux/mm.h index XXXXXXX..XXXXXXX 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -XXX,XX +XXX,XX @@ int get_user_pages_fast(unsigned long start, int nr_pages, unsigned int gup_flags, struct page **pages); int pin_user_pages_fast(unsigned long start, int nr_pages, unsigned int gup_flags, struct page **pages); +void folio_add_pin(struct folio *folio); int account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc); int __account_locked_vm(struct mm_struct *mm, unsigned long pages, bool inc, diff --git a/mm/gup.c b/mm/gup.c index XXXXXXX..XXXXXXX 100644 --- a/mm/gup.c +++ b/mm/gup.c @@ -XXX,XX +XXX,XX @@ void unpin_user_page(struct page *page) } EXPORT_SYMBOL(unpin_user_page); +/** + * folio_add_pin - Try to get an additional pin on a pinned folio + * @folio: The folio to be pinned + * + * Get an additional pin on a folio we already have a pin on. Makes no change + * if the folio is a zero_page. + */ +void folio_add_pin(struct folio *folio) +{ + if (is_zero_folio(folio)) + return; + + /* + * Similar to try_grab_folio(): be sure to *also* increment the normal + * page refcount field at least once, so that the page really is + * pinned. + */ + if (folio_test_large(folio)) { + WARN_ON_ONCE(atomic_read(&folio->_pincount) < 1); + folio_ref_add(folio, 1); + atomic_add(1, &folio->_pincount); + } else { + WARN_ON_ONCE(folio_ref_count(folio) < GUP_PIN_COUNTING_BIAS); + folio_ref_add(folio, GUP_PIN_COUNTING_BIAS); + } +} + static inline struct folio *gup_folio_range_next(struct page *start, unsigned long npages, unsigned long i, unsigned int *ntails) { Change the old block-based direct-I/O code to use iov_iter_extract_pages() to pin user pages or leave kernel pages unpinned rather than taking refs when submitting bios. This makes use of the preceding patches to not take pins on the zero page (thereby allowing insertion of zero pages in with pinned pages) and to get additional pins on pages, allowing an extracted page to be used in multiple bios without having to re-extract it. Signed-off-by: David Howells <dhowells@redhat.com> cc: Christoph Hellwig <hch@infradead.org> cc: David Hildenbrand <david@redhat.com> cc: Lorenzo Stoakes <lstoakes@gmail.com> cc: Andrew Morton <akpm@linux-foundation.org> cc: Jens Axboe <axboe@kernel.dk> cc: Al Viro <viro@zeniv.linux.org.uk> cc: Matthew Wilcox <willy@infradead.org> cc: Jan Kara <jack@suse.cz> cc: Jeff Layton <jlayton@kernel.org> cc: Jason Gunthorpe <jgg@nvidia.com> cc: Logan Gunthorpe <logang@deltatee.com> cc: Hillf Danton <hdanton@sina.com> cc: Christian Brauner <brauner@kernel.org> cc: Linus Torvalds <torvalds@linux-foundation.org> cc: linux-fsdevel@vger.kernel.org cc: linux-block@vger.kernel.org cc: linux-kernel@vger.kernel.org cc: linux-mm@kvack.org --- Notes: ver #3) - Rename need_unpin to is_pinned in struct dio. - page_get_additional_pin() was renamed to folio_add_pin(). ver #2) - Need to set BIO_PAGE_PINNED conditionally, not BIO_PAGE_REFFED. fs/direct-io.c | 72 ++++++++++++++++++++++++++++++-------------------- 1 file changed, 43 insertions(+), 29 deletions(-) diff --git a/fs/direct-io.c b/fs/direct-io.c index XXXXXXX..XXXXXXX 100644 --- a/fs/direct-io.c +++ b/fs/direct-io.c @@ -XXX,XX +XXX,XX @@ #include "internal.h" /* - * How many user pages to map in one call to get_user_pages(). This determines - * the size of a structure in the slab cache + * How many user pages to map in one call to iov_iter_extract_pages(). This + * determines the size of a structure in the slab cache */ #define DIO_PAGES 64 @@ -XXX,XX +XXX,XX @@ struct dio { struct inode *inode; loff_t i_size; /* i_size when submitted */ dio_iodone_t *end_io; /* IO completion function */ + bool is_pinned; /* T if we have pins on the pages */ void *private; /* copy from map_bh.b_private */ /* BIO completion state */ spinlock_t bio_lock; /* protects BIO fields below */ - int page_errors; /* errno from get_user_pages() */ + int page_errors; /* err from iov_iter_extract_pages() */ int is_async; /* is IO async ? */ bool defer_completion; /* defer AIO completion to workqueue? */ bool should_dirty; /* if pages should be dirtied */ @@ -XXX,XX +XXX,XX @@ static inline unsigned dio_pages_present(struct dio_submit *sdio) */ static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio) { + struct page **pages = dio->pages; const enum req_op dio_op = dio->opf & REQ_OP_MASK; ssize_t ret; - ret = iov_iter_get_pages2(sdio->iter, dio->pages, LONG_MAX, DIO_PAGES, - &sdio->from); + ret = iov_iter_extract_pages(sdio->iter, &pages, LONG_MAX, + DIO_PAGES, 0, &sdio->from); if (ret < 0 && sdio->blocks_available && dio_op == REQ_OP_WRITE) { - struct page *page = ZERO_PAGE(0); /* * A memory fault, but the filesystem has some outstanding * mapped blocks. We need to use those blocks up to avoid @@ -XXX,XX +XXX,XX @@ static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio) */ if (dio->page_errors == 0) dio->page_errors = ret; - get_page(page); - dio->pages[0] = page; + dio->pages[0] = ZERO_PAGE(0); sdio->head = 0; sdio->tail = 1; sdio->from = 0; @@ -XXX,XX +XXX,XX @@ static inline int dio_refill_pages(struct dio *dio, struct dio_submit *sdio) /* * Get another userspace page. Returns an ERR_PTR on error. Pages are - * buffered inside the dio so that we can call get_user_pages() against a - * decent number of pages, less frequently. To provide nicer use of the - * L1 cache. + * buffered inside the dio so that we can call iov_iter_extract_pages() + * against a decent number of pages, less frequently. To provide nicer use of + * the L1 cache. */ static inline struct page *dio_get_page(struct dio *dio, struct dio_submit *sdio) @@ -XXX,XX +XXX,XX @@ static inline struct page *dio_get_page(struct dio *dio, return dio->pages[sdio->head]; } +static void dio_pin_page(struct dio *dio, struct page *page) +{ + if (dio->is_pinned) + folio_add_pin(page_folio(page)); +} + +static void dio_unpin_page(struct dio *dio, struct page *page) +{ + if (dio->is_pinned) + unpin_user_page(page); +} + /* * dio_complete() - called when all DIO BIO I/O has been completed * @@ -XXX,XX +XXX,XX @@ dio_bio_alloc(struct dio *dio, struct dio_submit *sdio, bio->bi_end_io = dio_bio_end_aio; else bio->bi_end_io = dio_bio_end_io; - /* for now require references for all pages */ - bio_set_flag(bio, BIO_PAGE_REFFED); + if (dio->is_pinned) + bio_set_flag(bio, BIO_PAGE_PINNED); sdio->bio = bio; sdio->logical_offset_in_bio = sdio->cur_page_fs_offset; } @@ -XXX,XX +XXX,XX @@ static inline void dio_bio_submit(struct dio *dio, struct dio_submit *sdio) */ static inline void dio_cleanup(struct dio *dio, struct dio_submit *sdio) { - while (sdio->head < sdio->tail) - put_page(dio->pages[sdio->head++]); + if (dio->is_pinned) + unpin_user_pages(dio->pages + sdio->head, + sdio->tail - sdio->head); } /* @@ -XXX,XX +XXX,XX @@ static inline int dio_new_bio(struct dio *dio, struct dio_submit *sdio, * * Return zero on success. Non-zero means the caller needs to start a new BIO. */ -static inline int dio_bio_add_page(struct dio_submit *sdio) +static inline int dio_bio_add_page(struct dio *dio, struct dio_submit *sdio) { int ret; @@ -XXX,XX +XXX,XX @@ static inline int dio_bio_add_page(struct dio_submit *sdio) */ if ((sdio->cur_page_len + sdio->cur_page_offset) == PAGE_SIZE) sdio->pages_in_io--; - get_page(sdio->cur_page); + dio_pin_page(dio, sdio->cur_page); sdio->final_block_in_bio = sdio->cur_page_block + (sdio->cur_page_len >> sdio->blkbits); ret = 0; @@ -XXX,XX +XXX,XX @@ static inline int dio_send_cur_page(struct dio *dio, struct dio_submit *sdio, goto out; } - if (dio_bio_add_page(sdio) != 0) { + if (dio_bio_add_page(dio, sdio) != 0) { dio_bio_submit(dio, sdio); ret = dio_new_bio(dio, sdio, sdio->cur_page_block, map_bh); if (ret == 0) { - ret = dio_bio_add_page(sdio); + ret = dio_bio_add_page(dio, sdio); BUG_ON(ret != 0); } } @@ -XXX,XX +XXX,XX @@ submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page, */ if (sdio->cur_page) { ret = dio_send_cur_page(dio, sdio, map_bh); - put_page(sdio->cur_page); + dio_unpin_page(dio, sdio->cur_page); sdio->cur_page = NULL; if (ret) return ret; } - get_page(page); /* It is in dio */ + dio_pin_page(dio, page); /* It is in dio */ sdio->cur_page = page; sdio->cur_page_offset = offset; sdio->cur_page_len = len; @@ -XXX,XX +XXX,XX @@ submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page, ret = dio_send_cur_page(dio, sdio, map_bh); if (sdio->bio) dio_bio_submit(dio, sdio); - put_page(sdio->cur_page); + dio_unpin_page(dio, sdio->cur_page); sdio->cur_page = NULL; } return ret; @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, ret = get_more_blocks(dio, sdio, map_bh); if (ret) { - put_page(page); + dio_unpin_page(dio, page); goto out; } if (!buffer_mapped(map_bh)) @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, /* AKPM: eargh, -ENOTBLK is a hack */ if (dio_op == REQ_OP_WRITE) { - put_page(page); + dio_unpin_page(dio, page); return -ENOTBLK; } @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, if (sdio->block_in_file >= i_size_aligned >> blkbits) { /* We hit eof */ - put_page(page); + dio_unpin_page(dio, page); goto out; } zero_user(page, from, 1 << blkbits); @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, sdio->next_block_for_io, map_bh); if (ret) { - put_page(page); + dio_unpin_page(dio, page); goto out; } sdio->next_block_for_io += this_chunk_blocks; @@ -XXX,XX +XXX,XX @@ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, break; } - /* Drop the ref which was taken in get_user_pages() */ - put_page(page); + /* Drop the pin which was taken in get_user_pages() */ + dio_unpin_page(dio, page); } out: return ret; @@ -XXX,XX +XXX,XX @@ ssize_t __blockdev_direct_IO(struct kiocb *iocb, struct inode *inode, /* will be released by direct_io_worker */ inode_lock(inode); } + dio->is_pinned = iov_iter_extract_will_pin(iter); /* Once we sampled i_size check for reads beyond EOF */ dio->i_size = i_size_read(inode); @@ -XXX,XX +XXX,XX @@ ssize_t __blockdev_direct_IO(struct kiocb *iocb, struct inode *inode, ret2 = dio_send_cur_page(dio, &sdio, &map_bh); if (retval == 0) retval = ret2; - put_page(sdio.cur_page); + dio_unpin_page(dio, sdio.cur_page); sdio.cur_page = NULL; } if (sdio.bio)