| Author | Tokens | Token Proportion | Commits | Commit Proportion |
|---|---|---|---|---|
| Andrew Morton | 2145 | 46.28% | 30 | 16.39% |
| Andi Kleen | 641 | 13.83% | 7 | 3.83% |
| Christoph Hellwig | 364 | 7.85% | 30 | 16.39% |
| Al Viro | 182 | 3.93% | 12 | 6.56% |
| Zach Brown | 180 | 3.88% | 6 | 3.28% |
| David Howells | 156 | 3.37% | 4 | 2.19% |
| Bart Van Assche | 121 | 2.61% | 2 | 1.09% |
| Lukas Czerner | 100 | 2.16% | 2 | 1.09% |
| Jan Kara | 96 | 2.07% | 6 | 3.28% |
| Gabriel Krisman Bertazi | 78 | 1.68% | 3 | 1.64% |
| Josef Bacik | 49 | 1.06% | 2 | 1.09% |
| Omar Sandoval | 41 | 0.88% | 1 | 0.55% |
| Stephen Lord | 40 | 0.86% | 1 | 0.55% |
| Jeff Moyer | 39 | 0.84% | 3 | 1.64% |
| Goldwyn Rodrigues | 37 | 0.80% | 1 | 0.55% |
| Daniel McNeil | 33 | 0.71% | 2 | 1.09% |
| Michael Christie | 29 | 0.63% | 1 | 0.55% |
| Linus Torvalds | 28 | 0.60% | 4 | 2.19% |
| Nathan Scott | 27 | 0.58% | 3 | 1.64% |
| Linus Torvalds (pre-git) | 26 | 0.56% | 12 | 6.56% |
| Badari Pulavarty | 21 | 0.45% | 3 | 1.64% |
| Boaz Harrosh | 19 | 0.41% | 1 | 0.55% |
| Ming Lei | 18 | 0.39% | 2 | 1.09% |
| Fengguang Wu | 16 | 0.35% | 1 | 0.55% |
| Jens Axboe | 15 | 0.32% | 4 | 2.19% |
| Ernesto A. Fernández | 14 | 0.30% | 1 | 0.55% |
| Tao Ma | 12 | 0.26% | 1 | 0.55% |
| Chandan Rajendra | 11 | 0.24% | 1 | 0.55% |
| Jack Qiu | 9 | 0.19% | 1 | 0.55% |
| Kent Overstreet | 6 | 0.13% | 2 | 1.09% |
| Toshihiro Iwamoto | 6 | 0.13% | 1 | 0.55% |
| Eryu Guan | 6 | 0.13% | 2 | 1.09% |
| Nicholas Piggin | 6 | 0.13% | 2 | 1.09% |
| Andreas Gruenbacher | 6 | 0.13% | 1 | 0.55% |
| Kenneth W Chen | 5 | 0.11% | 1 | 0.55% |
| Maximilian Heyne | 5 | 0.11% | 1 | 0.55% |
| Theodore Y. Ts'o | 5 | 0.11% | 1 | 0.55% |
| Arnaldo Carvalho de Melo | 4 | 0.09% | 1 | 0.55% |
| Nikolay Borisov | 4 | 0.09% | 1 | 0.55% |
| Eric Sesterhenn / Snakebyte | 3 | 0.06% | 1 | 0.55% |
| Eric Biggers | 3 | 0.06% | 1 | 0.55% |
| Dave Jones | 3 | 0.06% | 1 | 0.55% |
| Randy Dunlap | 3 | 0.06% | 2 | 1.09% |
| Miklos Szeredi | 2 | 0.04% | 1 | 0.55% |
| Edward Shishkin | 2 | 0.04% | 1 | 0.55% |
| Julia Lawall | 2 | 0.04% | 1 | 0.55% |
| Sasha Levin | 2 | 0.04% | 1 | 0.55% |
| Neil Brown | 2 | 0.04% | 1 | 0.55% |
| Christoph Lameter | 1 | 0.02% | 1 | 0.55% |
| Jes Sorensen | 1 | 0.02% | 1 | 0.55% |
| Arun Sharma | 1 | 0.02% | 1 | 0.55% |
| Francois Cami | 1 | 0.02% | 1 | 0.55% |
| Matthew Wilcox | 1 | 0.02% | 1 | 0.55% |
| Robert P. J. Day | 1 | 0.02% | 1 | 0.55% |
| Thomas Gleixner | 1 | 0.02% | 1 | 0.55% |
| Alexey Dobriyan | 1 | 0.02% | 1 | 0.55% |
| Fabian Frederick | 1 | 0.02% | 1 | 0.55% |
| Abhinav Singh | 1 | 0.02% | 1 | 0.55% |
| Mark Rutland | 1 | 0.02% | 1 | 0.55% |
| Dave Dillow | 1 | 0.02% | 1 | 0.55% |
| Namhyung Kim | 1 | 0.02% | 1 | 0.55% |
| Total | 4635 | 183 |
// SPDX-License-Identifier: GPL-2.0-only /* * fs/direct-io.c * * Copyright (C) 2002, Linus Torvalds. * * O_DIRECT * * 04Jul2002 Andrew Morton * Initial version * 11Sep2002 janetinc@us.ibm.com * added readv/writev support. * 29Oct2002 Andrew Morton * rewrote bio_add_page() support. * 30Oct2002 pbadari@us.ibm.com * added support for non-aligned IO. * 06Nov2002 pbadari@us.ibm.com * added asynchronous IO support. * 21Jul2003 nathans@sgi.com * 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 64 /* * Flags for dio_complete() */ #define DIO_COMPLETE_ASYNC 0x01 /* This is async IO */ #define DIO_COMPLETE_INVALIDATE 0x02 /* Can invalidate pages */ /* * 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 { struct bio *bio; /* bio under assembly */ unsigned blkbits; /* doesn't change */ unsigned blkfactor; /* When we're using an alignment which is finer than the filesystem's soft blocksize, this specifies how much finer. blkfactor=2 means 1/4-block alignment. Does not change */ unsigned start_zero_done; /* flag: sub-blocksize zeroing has been performed at the start of a write */ int pages_in_io; /* approximate total IO pages */ sector_t block_in_file; /* Current offset into the underlying file in dio_block units. */ unsigned blocks_available; /* At block_in_file. changes */ int reap_counter; /* rate limit reaping */ sector_t final_block_in_request;/* doesn't change */ int boundary; /* prev block is at a boundary */ get_block_t *get_block; /* block mapping function */ loff_t logical_offset_in_bio; /* current first logical block in bio */ sector_t final_block_in_bio; /* current final block in bio + 1 */ sector_t next_block_for_io; /* next block to be put under IO, in dio_blocks units */ /* * Deferred addition of a page to the dio. These variables are * private to dio_send_cur_page(), submit_page_section() and * dio_bio_add_page(). */ struct page *cur_page; /* The page */ unsigned cur_page_offset; /* Offset into it, in bytes */ unsigned cur_page_len; /* Nr of bytes at cur_page_offset */ sector_t cur_page_block; /* Where it starts */ loff_t cur_page_fs_offset; /* Offset in file */ struct iov_iter *iter; /* * Page queue. These variables belong to dio_refill_pages() and * dio_get_page(). */ unsigned head; /* next page to process */ unsigned tail; /* last valid page + 1 */ size_t from, to; }; /* dio_state communicated between submission path and end_io */ struct dio { int flags; /* doesn't change */ blk_opf_t opf; /* request operation type and flags */ struct gendisk *bio_disk; 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; /* 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 */ int io_error; /* IO error in completion path */ unsigned long refcount; /* direct_io_worker() and bios */ struct bio *bio_list; /* singly linked via bi_private */ struct task_struct *waiter; /* waiting task (NULL if none) */ /* AIO related stuff */ struct kiocb *iocb; /* kiocb */ ssize_t result; /* IO result */ /* * pages[] (and any fields placed after it) are not zeroed out at * allocation time. Don't add new fields after pages[] unless you * wish that they not be zeroed. */ union { struct page *pages[DIO_PAGES]; /* page buffer */ struct work_struct complete_work;/* deferred AIO completion */ }; } ____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) { return sdio->tail - sdio->head; } /* * 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) { struct page **pages = dio->pages; const enum req_op dio_op = dio->opf & REQ_OP_MASK; ssize_t ret; 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) { /* * A memory fault, but the filesystem has some outstanding * mapped blocks. We need to use those blocks up to avoid * leaking stale data in the file. */ if (dio->page_errors == 0) dio->page_errors = ret; dio->pages[0] = ZERO_PAGE(0); sdio->head = 0; sdio->tail = 1; sdio->from = 0; sdio->to = PAGE_SIZE; return 0; } if (ret >= 0) { ret += sdio->from; sdio->head = 0; sdio->tail = (ret + PAGE_SIZE - 1) / PAGE_SIZE; sdio->to = ((ret - 1) & (PAGE_SIZE - 1)) + 1; return 0; } return ret; } /* * 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) { if (dio_pages_present(sdio) == 0) { int ret; ret = dio_refill_pages(dio, sdio); if (ret) return ERR_PTR(ret); BUG_ON(dio_pages_present(sdio) == 0); } 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 * * 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) { const enum req_op dio_op = dio->opf & REQ_OP_MASK; loff_t offset = dio->iocb->ki_pos; ssize_t transferred = 0; int err; /* * AIO submission can race with bio completion to get here while * expecting to have the last io completed by bio completion. * In that case -EIOCBQUEUED is in fact not an error we want * to preserve through this call. */ if (ret == -EIOCBQUEUED) ret = 0; if (dio->result) { transferred = dio->result; /* Check for short read case */ if (dio_op == REQ_OP_READ && ((offset + transferred) > dio->i_size)) transferred = dio->i_size - offset; /* ignore EFAULT if some IO has been done */ if (unlikely(ret == -EFAULT) && transferred) ret = 0; } if (ret == 0) ret = dio->page_errors; if (ret == 0) ret = dio->io_error; if (ret == 0) ret = transferred; if (dio->end_io) { // XXX: ki_pos?? err = dio->end_io(dio->iocb, offset, ret, dio->private); if (err) ret = err; } /* * Try again to invalidate clean pages which might have been cached by * non-direct readahead, or faulted in by get_user_pages() if the source * of the write was an mmap'ed region of the file we're writing. Either * one is a pretty crazy thing to do, so we don't support it 100%. If * this invalidation fails, tough, the write still worked... * * And this page cache invalidation has to be after dio->end_io(), as * some filesystems convert unwritten extents to real allocations in * end_io() when necessary, otherwise a racing buffer read would cache * zeros from unwritten extents. */ if (flags & DIO_COMPLETE_INVALIDATE && ret > 0 && dio_op == REQ_OP_WRITE) kiocb_invalidate_post_direct_write(dio->iocb, ret); inode_dio_end(dio->inode); if (flags & DIO_COMPLETE_ASYNC) { /* * generic_write_sync expects ki_pos to have been updated * already, but the submission path only does this for * synchronous I/O. */ dio->iocb->ki_pos += transferred; if (ret > 0 && dio_op == REQ_OP_WRITE) ret = generic_write_sync(dio->iocb, ret); dio->iocb->ki_complete(dio->iocb, ret); } kmem_cache_free(dio_cache, dio); return ret; } static void dio_aio_complete_work(struct work_struct *work) { struct dio *dio = container_of(work, struct dio, complete_work); dio_complete(dio, 0, DIO_COMPLETE_ASYNC | DIO_COMPLETE_INVALIDATE); } 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) { struct dio *dio = bio->bi_private; const enum req_op dio_op = dio->opf & REQ_OP_MASK; unsigned long remaining; unsigned long flags; bool defer_completion = false; /* cleanup the bio */ dio_bio_complete(dio, bio); spin_lock_irqsave(&dio->bio_lock, flags); remaining = --dio->refcount; if (remaining == 1 && dio->waiter) wake_up_process(dio->waiter); spin_unlock_irqrestore(&dio->bio_lock, flags); if (remaining == 0) { /* * Defer completion when defer_completion is set or * when the inode has pages mapped and this is AIO write. * We need to invalidate those pages because there is a * chance they contain stale data in the case buffered IO * went in between AIO submission and completion into the * same region. */ if (dio->result) defer_completion = dio->defer_completion || (dio_op == REQ_OP_WRITE && dio->inode->i_mapping->nrpages); if (defer_completion) { INIT_WORK(&dio->complete_work, dio_aio_complete_work); queue_work(dio->inode->i_sb->s_dio_done_wq, &dio->complete_work); } else { dio_complete(dio, 0, DIO_COMPLETE_ASYNC); } } } /* * 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) { struct dio *dio = bio->bi_private; unsigned long flags; spin_lock_irqsave(&dio->bio_lock, flags); bio->bi_private = dio->bio_list; dio->bio_list = bio; if (--dio->refcount == 1 && dio->waiter) wake_up_process(dio->waiter); spin_unlock_irqrestore(&dio->bio_lock, flags); } static inline void dio_bio_alloc(struct dio *dio, struct dio_submit *sdio, struct block_device *bdev, sector_t first_sector, int nr_vecs) { struct bio *bio; /* * bio_alloc() is guaranteed to return a bio when allowed to sleep and * we request a valid number of vectors. */ bio = bio_alloc(bdev, nr_vecs, dio->opf, GFP_KERNEL); bio->bi_iter.bi_sector = first_sector; if (dio->is_async) bio->bi_end_io = dio_bio_end_aio; else bio->bi_end_io = dio_bio_end_io; if (dio->is_pinned) bio_set_flag(bio, BIO_PAGE_PINNED); bio->bi_write_hint = file_inode(dio->iocb->ki_filp)->i_write_hint; sdio->bio = bio; sdio->logical_offset_in_bio = sdio->cur_page_fs_offset; } /* * 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) { const enum req_op dio_op = dio->opf & REQ_OP_MASK; struct bio *bio = sdio->bio; unsigned long flags; bio->bi_private = dio; spin_lock_irqsave(&dio->bio_lock, flags); dio->refcount++; spin_unlock_irqrestore(&dio->bio_lock, flags); if (dio->is_async && dio_op == REQ_OP_READ && dio->should_dirty) bio_set_pages_dirty(bio); dio->bio_disk = bio->bi_bdev->bd_disk; submit_bio(bio); sdio->bio = NULL; sdio->boundary = 0; sdio->logical_offset_in_bio = 0; } /* * Release any resources in case of a failure */ static inline void dio_cleanup(struct dio *dio, struct dio_submit *sdio) { if (dio->is_pinned) unpin_user_pages(dio->pages + sdio->head, sdio->tail - sdio->head); sdio->head = sdio->tail; } /* * 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) { unsigned long flags; struct bio *bio = NULL; spin_lock_irqsave(&dio->bio_lock, flags); /* * Wait as long as the list is empty and there are bios in flight. bio * completion drops the count, maybe adds to the list, and wakes while * holding the bio_lock so we don't need set_current_state()'s barrier * and can call it after testing our condition. */ while (dio->refcount > 1 && dio->bio_list == NULL) { __set_current_state(TASK_UNINTERRUPTIBLE); dio->waiter = current; spin_unlock_irqrestore(&dio->bio_lock, flags); blk_io_schedule(); /* wake up sets us TASK_RUNNING */ spin_lock_irqsave(&dio->bio_lock, flags); dio->waiter = NULL; } if (dio->bio_list) { bio = dio->bio_list; dio->bio_list = bio->bi_private; } spin_unlock_irqrestore(&dio->bio_lock, flags); return bio; } /* * Process one completed BIO. No locks are held. */ static blk_status_t dio_bio_complete(struct dio *dio, struct bio *bio) { blk_status_t err = bio->bi_status; const enum req_op dio_op = dio->opf & REQ_OP_MASK; bool should_dirty = dio_op == REQ_OP_READ && dio->should_dirty; if (err) { if (err == BLK_STS_AGAIN && (bio->bi_opf & REQ_NOWAIT)) dio->io_error = -EAGAIN; else dio->io_error = -EIO; } if (dio->is_async && should_dirty) { bio_check_pages_dirty(bio); /* transfers ownership */ } else { bio_release_pages(bio, should_dirty); bio_put(bio); } return err; } /* * 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) { struct bio *bio; do { bio = dio_await_one(dio); if (bio) dio_bio_complete(dio, bio); } while (bio); } /* * 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) { int ret = 0; if (sdio->reap_counter++ >= 64) { while (dio->bio_list) { unsigned long flags; struct bio *bio; int ret2; spin_lock_irqsave(&dio->bio_lock, flags); bio = dio->bio_list; dio->bio_list = bio->bi_private; spin_unlock_irqrestore(&dio->bio_lock, flags); ret2 = blk_status_to_errno(dio_bio_complete(dio, bio)); if (ret == 0) ret = ret2; } sdio->reap_counter = 0; } return ret; } static int dio_set_defer_completion(struct dio *dio) { struct super_block *sb = dio->inode->i_sb; if (dio->defer_completion) return 0; dio->defer_completion = true; if (!sb->s_dio_done_wq) return sb_init_dio_done_wq(sb); return 0; } /* * 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) { const enum req_op dio_op = dio->opf & REQ_OP_MASK; int ret; sector_t fs_startblk; /* Into file, in filesystem-sized blocks */ sector_t fs_endblk; /* Into file, in filesystem-sized blocks */ unsigned long fs_count; /* Number of filesystem-sized blocks */ int create; unsigned int i_blkbits = sdio->blkbits + sdio->blkfactor; loff_t i_size; /* * If there was a memory error and we've overwritten all the * mapped blocks then we can now return that memory error */ ret = dio->page_errors; if (ret == 0) { BUG_ON(sdio->block_in_file >= sdio->final_block_in_request); fs_startblk = sdio->block_in_file >> sdio->blkfactor; fs_endblk = (sdio->final_block_in_request - 1) >> sdio->blkfactor; fs_count = fs_endblk - fs_startblk + 1; map_bh->b_state = 0; map_bh->b_size = fs_count << i_blkbits; /* * For writes that could fill holes inside i_size on a * DIO_SKIP_HOLES filesystem we forbid block creations: only * overwrites are permitted. We will return early to the caller * once we see an unmapped buffer head returned, and the caller * will fall back to buffered I/O. * * Otherwise the decision is left to the get_blocks method, * which may decide to handle it or also return an unmapped * buffer head. */ create = dio_op == REQ_OP_WRITE; if (dio->flags & DIO_SKIP_HOLES) { i_size = i_size_read(dio->inode); if (i_size && fs_startblk <= (i_size - 1) >> i_blkbits) create = 0; } ret = (*sdio->get_block)(dio->inode, fs_startblk, map_bh, create); /* Store for completion */ dio->private = map_bh->b_private; if (ret == 0 && buffer_defer_completion(map_bh)) ret = dio_set_defer_completion(dio); } return ret; } /* * 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) { sector_t sector; int ret, nr_pages; ret = dio_bio_reap(dio, sdio); if (ret) goto out; sector = start_sector << (sdio->blkbits - 9); nr_pages = bio_max_segs(sdio->pages_in_io); BUG_ON(nr_pages <= 0); dio_bio_alloc(dio, sdio, map_bh->b_bdev, sector, nr_pages); sdio->boundary = 0; out: return ret; } /* * 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) { int ret; ret = bio_add_page(sdio->bio, sdio->cur_page, sdio->cur_page_len, sdio->cur_page_offset); if (ret == sdio->cur_page_len) { /* * Decrement count only, if we are done with this page */ if ((sdio->cur_page_len + sdio->cur_page_offset) == PAGE_SIZE) sdio->pages_in_io--; dio_pin_page(dio, sdio->cur_page); sdio->final_block_in_bio = sdio->cur_page_block + (sdio->cur_page_len >> sdio->blkbits); ret = 0; } else { ret = 1; } return ret; } /* * 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) { int ret = 0; if (sdio->bio) { loff_t cur_offset = sdio->cur_page_fs_offset; loff_t bio_next_offset = sdio->logical_offset_in_bio + sdio->bio->bi_iter.bi_size; /* * See whether this new request is contiguous with the old. * * Btrfs cannot handle having logically non-contiguous requests * submitted. For example if you have * * Logical: [0-4095][HOLE][8192-12287] * Physical: [0-4095] [4096-8191] * * We cannot submit those pages together as one BIO. So if our * current logical offset in the file does not equal what would * be the next logical offset in the bio, submit the bio we * have. */ if (sdio->final_block_in_bio != sdio->cur_page_block || cur_offset != bio_next_offset) dio_bio_submit(dio, sdio); } if (sdio->bio == NULL) { ret = dio_new_bio(dio, sdio, sdio->cur_page_block, map_bh); if (ret) goto out; } 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(dio, sdio); BUG_ON(ret != 0); } } out: return ret; } /* * 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) { const enum req_op dio_op = dio->opf & REQ_OP_MASK; int ret = 0; int boundary = sdio->boundary; /* dio_send_cur_page may clear it */ if (dio_op == REQ_OP_WRITE) { /* * Read accounting is performed in submit_bio() */ task_io_account_write(len); } /* * Can we just grow the current page's presence in the dio? */ if (sdio->cur_page == page && sdio->cur_page_offset + sdio->cur_page_len == offset && sdio->cur_page_block + (sdio->cur_page_len >> sdio->blkbits) == blocknr) { sdio->cur_page_len += len; goto out; } /* * If there's a deferred page already there then send it. */ if (sdio->cur_page) { ret = dio_send_cur_page(dio, sdio, map_bh); dio_unpin_page(dio, sdio->cur_page); sdio->cur_page = NULL; if (ret) return ret; } dio_pin_page(dio, page); /* It is in dio */ sdio->cur_page = page; sdio->cur_page_offset = offset; sdio->cur_page_len = len; sdio->cur_page_block = blocknr; sdio->cur_page_fs_offset = sdio->block_in_file << sdio->blkbits; out: /* * If boundary then we want to schedule the IO now to * avoid metadata seeks. */ if (boundary) { ret = dio_send_cur_page(dio, sdio, map_bh); if (sdio->bio) dio_bio_submit(dio, sdio); dio_unpin_page(dio, sdio->cur_page); sdio->cur_page = NULL; } return ret; } /* * 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) { unsigned dio_blocks_per_fs_block; unsigned this_chunk_blocks; /* In dio_blocks */ unsigned this_chunk_bytes; struct page *page; sdio->start_zero_done = 1; if (!sdio->blkfactor || !buffer_new(map_bh)) return; dio_blocks_per_fs_block = 1 << sdio->blkfactor; this_chunk_blocks = sdio->block_in_file & (dio_blocks_per_fs_block - 1); if (!this_chunk_blocks) return; /* * We need to zero out part of an fs block. It is either at the * beginning or the end of the fs block. */ if (end) this_chunk_blocks = dio_blocks_per_fs_block - this_chunk_blocks; this_chunk_bytes = this_chunk_blocks << sdio->blkbits; page = ZERO_PAGE(0); if (submit_page_section(dio, sdio, page, 0, this_chunk_bytes, sdio->next_block_for_io, map_bh)) return; sdio->next_block_for_io += this_chunk_blocks; } /* * 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) { const enum req_op dio_op = dio->opf & REQ_OP_MASK; const unsigned blkbits = sdio->blkbits; const unsigned i_blkbits = blkbits + sdio->blkfactor; int ret = 0; while (sdio->block_in_file < sdio->final_block_in_request) { struct page *page; size_t from, to; page = dio_get_page(dio, sdio); if (IS_ERR(page)) { ret = PTR_ERR(page); goto out; } from = sdio->head ? 0 : sdio->from; to = (sdio->head == sdio->tail - 1) ? sdio->to : PAGE_SIZE; sdio->head++; while (from < to) { unsigned this_chunk_bytes; /* # of bytes mapped */ unsigned this_chunk_blocks; /* # of blocks */ unsigned u; if (sdio->blocks_available == 0) { /* * Need to go and map some more disk */ unsigned long blkmask; unsigned long dio_remainder; ret = get_more_blocks(dio, sdio, map_bh); if (ret) { dio_unpin_page(dio, page); goto out; } if (!buffer_mapped(map_bh)) goto do_holes; sdio->blocks_available = map_bh->b_size >> blkbits; sdio->next_block_for_io = map_bh->b_blocknr << sdio->blkfactor; if (buffer_new(map_bh)) { clean_bdev_aliases( map_bh->b_bdev, map_bh->b_blocknr, map_bh->b_size >> i_blkbits); } if (!sdio->blkfactor) goto do_holes; blkmask = (1 << sdio->blkfactor) - 1; dio_remainder = (sdio->block_in_file & blkmask); /* * If we are at the start of IO and that IO * starts partway into a fs-block, * dio_remainder will be non-zero. If the IO * is a read then we can simply advance the IO * cursor to the first block which is to be * read. But if the IO is a write and the * block was newly allocated we cannot do that; * the start of the fs block must be zeroed out * on-disk */ if (!buffer_new(map_bh)) sdio->next_block_for_io += dio_remainder; sdio->blocks_available -= dio_remainder; } do_holes: /* Handle holes */ if (!buffer_mapped(map_bh)) { loff_t i_size_aligned; /* AKPM: eargh, -ENOTBLK is a hack */ if (dio_op == REQ_OP_WRITE) { dio_unpin_page(dio, page); return -ENOTBLK; } /* * Be sure to account for a partial block as the * last block in the file */ i_size_aligned = ALIGN(i_size_read(dio->inode), 1 << blkbits); if (sdio->block_in_file >= i_size_aligned >> blkbits) { /* We hit eof */ dio_unpin_page(dio, page); goto out; } zero_user(page, from, 1 << blkbits); sdio->block_in_file++; from += 1 << blkbits; dio->result += 1 << blkbits; goto next_block; } /* * If we're performing IO which has an alignment which * is finer than the underlying fs, go check to see if * we must zero out the start of this block. */ if (unlikely(sdio->blkfactor && !sdio->start_zero_done)) dio_zero_block(dio, sdio, 0, map_bh); /* * Work out, in this_chunk_blocks, how much disk we * can add to this page */ this_chunk_blocks = sdio->blocks_available; u = (to - from) >> blkbits; if (this_chunk_blocks > u) this_chunk_blocks = u; u = sdio->final_block_in_request - sdio->block_in_file; if (this_chunk_blocks > u) this_chunk_blocks = u; this_chunk_bytes = this_chunk_blocks << blkbits; BUG_ON(this_chunk_bytes == 0); if (this_chunk_blocks == sdio->blocks_available) sdio->boundary = buffer_boundary(map_bh); ret = submit_page_section(dio, sdio, page, from, this_chunk_bytes, sdio->next_block_for_io, map_bh); if (ret) { dio_unpin_page(dio, page); goto out; } sdio->next_block_for_io += this_chunk_blocks; sdio->block_in_file += this_chunk_blocks; from += this_chunk_bytes; dio->result += this_chunk_bytes; sdio->blocks_available -= this_chunk_blocks; next_block: BUG_ON(sdio->block_in_file > sdio->final_block_in_request); if (sdio->block_in_file == sdio->final_block_in_request) break; } /* Drop the pin which was taken in get_user_pages() */ dio_unpin_page(dio, page); } out: return ret; } static inline int drop_refcount(struct dio *dio) { int ret2; unsigned long flags; /* * Sync will always be dropping the final ref and completing the * operation. AIO can if it was a broken operation described above or * in fact if all the bios race to complete before we get here. In * that case dio_complete() translates the EIOCBQUEUED into the proper * return code that the caller will hand to ->complete(). * * This is managed by the bio_lock instead of being an atomic_t so that * completion paths can drop their ref and use the remaining count to * decide to wake the submission path atomically. */ spin_lock_irqsave(&dio->bio_lock, flags); ret2 = --dio->refcount; spin_unlock_irqrestore(&dio->bio_lock, flags); return ret2; } /* * 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) { unsigned i_blkbits = READ_ONCE(inode->i_blkbits); unsigned blkbits = i_blkbits; unsigned blocksize_mask = (1 << blkbits) - 1; ssize_t retval = -EINVAL; const size_t count = iov_iter_count(iter); loff_t offset = iocb->ki_pos; const loff_t end = offset + count; struct dio *dio; struct dio_submit sdio = { NULL, }; struct buffer_head map_bh = { 0, }; struct blk_plug plug; unsigned long align = offset | iov_iter_alignment(iter); /* * Avoid references to bdev if not absolutely needed to give * the early prefetch in the caller enough time. */ /* watch out for a 0 len io from a tricksy fs */ if (iov_iter_rw(iter) == READ && !count) return 0; dio = kmem_cache_alloc(dio_cache, GFP_KERNEL); if (!dio) return -ENOMEM; /* * Believe it or not, zeroing out the page array caused a .5% * performance regression in a database benchmark. So, we take * care to only zero out what's needed. */ memset(dio, 0, offsetof(struct dio, pages)); dio->flags = flags; if (dio->flags & DIO_LOCKING && iov_iter_rw(iter) == READ) { /* 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); if (iov_iter_rw(iter) == READ && offset >= dio->i_size) { retval = 0; goto fail_dio; } if (align & blocksize_mask) { if (bdev) blkbits = blksize_bits(bdev_logical_block_size(bdev)); blocksize_mask = (1 << blkbits) - 1; if (align & blocksize_mask) goto fail_dio; } if (dio->flags & DIO_LOCKING && iov_iter_rw(iter) == READ) { struct address_space *mapping = iocb->ki_filp->f_mapping; retval = filemap_write_and_wait_range(mapping, offset, end - 1); if (retval) goto fail_dio; } /* * For file extending writes updating i_size before data writeouts * complete can expose uninitialized blocks in dumb filesystems. * In that case we need to wait for I/O completion even if asked * for an asynchronous write. */ if (is_sync_kiocb(iocb)) dio->is_async = false; else if (iov_iter_rw(iter) == WRITE && end > i_size_read(inode)) dio->is_async = false; else dio->is_async = true; dio->inode = inode; if (iov_iter_rw(iter) == WRITE) { dio->opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; if (iocb->ki_flags & IOCB_NOWAIT) dio->opf |= REQ_NOWAIT; } else { dio->opf = REQ_OP_READ; } /* * For AIO O_(D)SYNC writes we need to defer completions to a workqueue * so that we can call ->fsync. */ if (dio->is_async && iov_iter_rw(iter) == WRITE) { retval = 0; if (iocb_is_dsync(iocb)) retval = dio_set_defer_completion(dio); else if (!dio->inode->i_sb->s_dio_done_wq) { /* * In case of AIO write racing with buffered read we * need to defer completion. We can't decide this now, * however the workqueue needs to be initialized here. */ retval = sb_init_dio_done_wq(dio->inode->i_sb); } if (retval) goto fail_dio; } /* * Will be decremented at I/O completion time. */ inode_dio_begin(inode); sdio.blkbits = blkbits; sdio.blkfactor = i_blkbits - blkbits; sdio.block_in_file = offset >> blkbits; sdio.get_block = get_block; dio->end_io = end_io; sdio.final_block_in_bio = -1; sdio.next_block_for_io = -1; dio->iocb = iocb; spin_lock_init(&dio->bio_lock); dio->refcount = 1; dio->should_dirty = user_backed_iter(iter) && iov_iter_rw(iter) == READ; sdio.iter = iter; sdio.final_block_in_request = end >> blkbits; /* * In case of non-aligned buffers, we may need 2 more * pages since we need to zero out first and last block. */ if (unlikely(sdio.blkfactor)) sdio.pages_in_io = 2; sdio.pages_in_io += iov_iter_npages(iter, INT_MAX); blk_start_plug(&plug); retval = do_direct_IO(dio, &sdio, &map_bh); if (retval) dio_cleanup(dio, &sdio); if (retval == -ENOTBLK) { /* * The remaining part of the request will be * handled by buffered I/O when we return */ retval = 0; } /* * There may be some unwritten disk at the end of a part-written * fs-block-sized block. Go zero that now. */ dio_zero_block(dio, &sdio, 1, &map_bh); if (sdio.cur_page) { ssize_t ret2; ret2 = dio_send_cur_page(dio, &sdio, &map_bh); if (retval == 0) retval = ret2; dio_unpin_page(dio, sdio.cur_page); sdio.cur_page = NULL; } if (sdio.bio) dio_bio_submit(dio, &sdio); blk_finish_plug(&plug); /* * It is possible that, we return short IO due to end of file. * In that case, we need to release all the pages we got hold on. */ dio_cleanup(dio, &sdio); /* * All block lookups have been performed. For READ requests * we can let i_mutex go now that its achieved its purpose * of protecting us from looking up uninitialized blocks. */ if (iov_iter_rw(iter) == READ && (dio->flags & DIO_LOCKING)) inode_unlock(dio->inode); /* * The only time we want to leave bios in flight is when a successful * partial aio read or full aio write have been setup. In that case * bio completion will call aio_complete. The only time it's safe to * call aio_complete is when we return -EIOCBQUEUED, so we key on that. * This had *better* be the only place that raises -EIOCBQUEUED. */ BUG_ON(retval == -EIOCBQUEUED); if (dio->is_async && retval == 0 && dio->result && (iov_iter_rw(iter) == READ || dio->result == count)) retval = -EIOCBQUEUED; else dio_await_completion(dio); if (drop_refcount(dio) == 0) { retval = dio_complete(dio, retval, DIO_COMPLETE_INVALIDATE); } else BUG_ON(retval != -EIOCBQUEUED); return retval; fail_dio: if (dio->flags & DIO_LOCKING && iov_iter_rw(iter) == READ) inode_unlock(inode); kmem_cache_free(dio_cache, dio); return retval; } EXPORT_SYMBOL(__blockdev_direct_IO); static __init int dio_init(void) { dio_cache = KMEM_CACHE(dio, SLAB_PANIC); return 0; } module_init(dio_init)
Information contained on this website is for historical information purposes only and does not indicate or represent copyright ownership.
Created with Cregit http://github.com/cregit/cregit
Version 2.0-RC1