// SPDX-License-Identifier: GPL-2.0-only /* * fs/direct-io.c * * Copyright (C) 2002, Linus Torvalds. * * O_DIRECT * * 04Jul2002 Andrew Morton * Initial version * 11Sep2002 [email protected] * added readv/writev support. * 29Oct2002 Andrew Morton * rewrote bio_add_page() support. * 30Oct2002 [email protected] * added support for non-aligned IO. * 06Nov2002 [email protected] * added asynchronous IO support. * 21Jul2003 [email protected] * added IO completion notifier. */ #include <linux/kernel.h> #include <linux/module.h> #include <linux/types.h> #include <linux/fs.h> #include <linux/mm.h> #include <linux/slab.h> #include <linux/highmem.h> #include <linux/pagemap.h> #include <linux/task_io_accounting_ops.h> #include <linux/bio.h> #include <linux/wait.h> #include <linux/err.h> #include <linux/blkdev.h> #include <linux/buffer_head.h> #include <linux/rwsem.h> #include <linux/uio.h> #include <linux/atomic.h> #include <linux/prefetch.h> #include "internal.h" /* * 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 … /* * Flags for dio_complete() */ #define DIO_COMPLETE_ASYNC … #define DIO_COMPLETE_INVALIDATE … /* * This code generally works in units of "dio_blocks". A dio_block is * somewhere between the hard sector size and the filesystem block size. it * is determined on a per-invocation basis. When talking to the filesystem * we need to convert dio_blocks to fs_blocks by scaling the dio_block quantity * down by dio->blkfactor. Similarly, fs-blocksize quantities are converted * to bio_block quantities by shifting left by blkfactor. * * If blkfactor is zero then the user's request was aligned to the filesystem's * blocksize. */ /* dio_state only used in the submission path */ struct dio_submit { … }; /* dio_state communicated between submission path and end_io */ struct dio { … } ____cacheline_aligned_in_smp; static struct kmem_cache *dio_cache __ro_after_init; /* * How many pages are in the queue? */ static inline unsigned dio_pages_present(struct dio_submit *sdio) { … } /* * Go grab and pin some userspace pages. Typically we'll get 64 at a time. */ 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 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) { … } static void dio_pin_page(struct dio *dio, struct page *page) { … } static void dio_unpin_page(struct dio *dio, struct page *page) { … } /* * dio_complete() - called when all DIO BIO I/O has been completed * * This drops i_dio_count, lets interested parties know that a DIO operation * has completed, and calculates the resulting return code for the operation. * * It lets the filesystem know if it registered an interest earlier via * get_block. Pass the private field of the map buffer_head so that * filesystems can use it to hold additional state between get_block calls and * dio_complete. */ static ssize_t dio_complete(struct dio *dio, ssize_t ret, unsigned int flags) { … } static void dio_aio_complete_work(struct work_struct *work) { … } static blk_status_t dio_bio_complete(struct dio *dio, struct bio *bio); /* * Asynchronous IO callback. */ static void dio_bio_end_aio(struct bio *bio) { … } /* * The BIO completion handler simply queues the BIO up for the process-context * handler. * * During I/O bi_private points at the dio. After I/O, bi_private is used to * implement a singly-linked list of completed BIOs, at dio->bio_list. */ static void dio_bio_end_io(struct bio *bio) { … } static inline void dio_bio_alloc(struct dio *dio, struct dio_submit *sdio, struct block_device *bdev, sector_t first_sector, int nr_vecs) { … } /* * In the AIO read case we speculatively dirty the pages before starting IO. * During IO completion, any of these pages which happen to have been written * back will be redirtied by bio_check_pages_dirty(). * * bios hold a dio reference between submit_bio and ->end_io. */ static inline void dio_bio_submit(struct dio *dio, struct dio_submit *sdio) { … } /* * Release any resources in case of a failure */ static inline void dio_cleanup(struct dio *dio, struct dio_submit *sdio) { … } /* * Wait for the next BIO to complete. Remove it and return it. NULL is * returned once all BIOs have been completed. This must only be called once * all bios have been issued so that dio->refcount can only decrease. This * requires that the caller hold a reference on the dio. */ static struct bio *dio_await_one(struct dio *dio) { … } /* * Process one completed BIO. No locks are held. */ static blk_status_t dio_bio_complete(struct dio *dio, struct bio *bio) { … } /* * Wait on and process all in-flight BIOs. This must only be called once * all bios have been issued so that the refcount can only decrease. * This just waits for all bios to make it through dio_bio_complete. IO * errors are propagated through dio->io_error and should be propagated via * dio_complete(). */ static void dio_await_completion(struct dio *dio) { … } /* * A really large O_DIRECT read or write can generate a lot of BIOs. So * to keep the memory consumption sane we periodically reap any completed BIOs * during the BIO generation phase. * * This also helps to limit the peak amount of pinned userspace memory. */ static inline int dio_bio_reap(struct dio *dio, struct dio_submit *sdio) { … } static int dio_set_defer_completion(struct dio *dio) { … } /* * Call into the fs to map some more disk blocks. We record the current number * of available blocks at sdio->blocks_available. These are in units of the * fs blocksize, i_blocksize(inode). * * The fs is allowed to map lots of blocks at once. If it wants to do that, * it uses the passed inode-relative block number as the file offset, as usual. * * get_block() is passed the number of i_blkbits-sized blocks which direct_io * has remaining to do. The fs should not map more than this number of blocks. * * If the fs has mapped a lot of blocks, it should populate bh->b_size to * indicate how much contiguous disk space has been made available at * bh->b_blocknr. * * If *any* of the mapped blocks are new, then the fs must set buffer_new(). * This isn't very efficient... * * In the case of filesystem holes: the fs may return an arbitrarily-large * hole by returning an appropriate value in b_size and by clearing * buffer_mapped(). However the direct-io code will only process holes one * block at a time - it will repeatedly call get_block() as it walks the hole. */ static int get_more_blocks(struct dio *dio, struct dio_submit *sdio, struct buffer_head *map_bh) { … } /* * There is no bio. Make one now. */ static inline int dio_new_bio(struct dio *dio, struct dio_submit *sdio, sector_t start_sector, struct buffer_head *map_bh) { … } /* * Attempt to put the current chunk of 'cur_page' into the current BIO. If * that was successful then update final_block_in_bio and take a ref against * the just-added page. * * Return zero on success. Non-zero means the caller needs to start a new BIO. */ static inline int dio_bio_add_page(struct dio *dio, struct dio_submit *sdio) { … } /* * Put cur_page under IO. The section of cur_page which is described by * cur_page_offset,cur_page_len is put into a BIO. The section of cur_page * starts on-disk at cur_page_block. * * We take a ref against the page here (on behalf of its presence in the bio). * * The caller of this function is responsible for removing cur_page from the * dio, and for dropping the refcount which came from that presence. */ static inline int dio_send_cur_page(struct dio *dio, struct dio_submit *sdio, struct buffer_head *map_bh) { … } /* * An autonomous function to put a chunk of a page under deferred IO. * * The caller doesn't actually know (or care) whether this piece of page is in * a BIO, or is under IO or whatever. We just take care of all possible * situations here. The separation between the logic of do_direct_IO() and * that of submit_page_section() is important for clarity. Please don't break. * * The chunk of page starts on-disk at blocknr. * * We perform deferred IO, by recording the last-submitted page inside our * private part of the dio structure. If possible, we just expand the IO * across that page here. * * If that doesn't work out then we put the old page into the bio and add this * page to the dio instead. */ static inline int submit_page_section(struct dio *dio, struct dio_submit *sdio, struct page *page, unsigned offset, unsigned len, sector_t blocknr, struct buffer_head *map_bh) { … } /* * If we are not writing the entire block and get_block() allocated * the block for us, we need to fill-in the unused portion of the * block with zeros. This happens only if user-buffer, fileoffset or * io length is not filesystem block-size multiple. * * `end' is zero if we're doing the start of the IO, 1 at the end of the * IO. */ static inline void dio_zero_block(struct dio *dio, struct dio_submit *sdio, int end, struct buffer_head *map_bh) { … } /* * Walk the user pages, and the file, mapping blocks to disk and generating * a sequence of (page,offset,len,block) mappings. These mappings are injected * into submit_page_section(), which takes care of the next stage of submission * * Direct IO against a blockdev is different from a file. Because we can * happily perform page-sized but 512-byte aligned IOs. It is important that * blockdev IO be able to have fine alignment and large sizes. * * So what we do is to permit the ->get_block function to populate bh.b_size * with the size of IO which is permitted at this offset and this i_blkbits. * * For best results, the blockdev should be set up with 512-byte i_blkbits and * it should set b_size to PAGE_SIZE or more inside get_block(). This gives * fine alignment but still allows this function to work in PAGE_SIZE units. */ static int do_direct_IO(struct dio *dio, struct dio_submit *sdio, struct buffer_head *map_bh) { … } static inline int drop_refcount(struct dio *dio) { … } /* * This is a library function for use by filesystem drivers. * * The locking rules are governed by the flags parameter: * - if the flags value contains DIO_LOCKING we use a fancy locking * scheme for dumb filesystems. * For writes this function is called under i_mutex and returns with * i_mutex held, for reads, i_mutex is not held on entry, but it is * taken and dropped again before returning. * - if the flags value does NOT contain DIO_LOCKING we don't use any * internal locking but rather rely on the filesystem to synchronize * direct I/O reads/writes versus each other and truncate. * * To help with locking against truncate we incremented the i_dio_count * counter before starting direct I/O, and decrement it once we are done. * Truncate can wait for it to reach zero to provide exclusion. It is * expected that filesystem provide exclusion between new direct I/O * and truncates. For DIO_LOCKING filesystems this is done by i_mutex, * but other filesystems need to take care of this on their own. * * NOTE: if you pass "sdio" to anything by pointer make sure that function * is always inlined. Otherwise gcc is unable to split the structure into * individual fields and will generate much worse code. This is important * for the whole file. */ ssize_t __blockdev_direct_IO(struct kiocb *iocb, struct inode *inode, struct block_device *bdev, struct iov_iter *iter, get_block_t get_block, dio_iodone_t end_io, int flags) { … } EXPORT_SYMBOL(…); static __init int dio_init(void) { … } module_init(…) …
Codebase Browser
linux