From e0bd70c67bf996b360f706b6c643000f2e384681 Mon Sep 17 00:00:00 2001 From: Filipe Manana Date: Wed, 27 Jan 2016 10:20:58 +0000 Subject: [PATCH 1/4] Btrfs: fix invalid page accesses in extent_same (dedup) ioctl In the extent_same ioctl we are getting the pages for the source and target ranges and unlocking them immediately after, which is incorrect because later we attempt to map them (with kmap_atomic) and access their contents at btrfs_cmp_data(). When we do such access the pages might have been relocated or removed from memory, which leads to an invalid memory access. This issue is detected on a kernel with CONFIG_DEBUG_PAGEALLOC=y which produces a trace like the following: 186736.677437] general protection fault: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC [186736.680382] Modules linked in: btrfs dm_flakey dm_mod ppdev xor raid6_pq sha256_generic hmac drbg ansi_cprng acpi_cpufreq evdev sg aesni_intel aes_x86_64 parport_pc ablk_helper tpm_tis psmouse parport i2c_piix4 tpm cryptd i2c_core lrw processor button serio_raw pcspkr gf128mul glue_helper loop autofs4 ext4 crc16 mbcache jbd2 sd_mod sr_mod cdrom ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring crc32c_intel scsi_mod e1000 virtio floppy [last unloaded: btrfs] [186736.681319] CPU: 13 PID: 10222 Comm: duperemove Tainted: G W 4.4.0-rc6-btrfs-next-18+ #1 [186736.681319] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS by qemu-project.org 04/01/2014 [186736.681319] task: ffff880132600400 ti: ffff880362284000 task.ti: ffff880362284000 [186736.681319] RIP: 0010:[] [] memcmp+0xb/0x22 [186736.681319] RSP: 0018:ffff880362287d70 EFLAGS: 00010287 [186736.681319] RAX: 000002c002468acf RBX: 0000000012345678 RCX: 0000000000000000 [186736.681319] RDX: 0000000000001000 RSI: 0005d129c5cf9000 RDI: 0005d129c5cf9000 [186736.681319] RBP: ffff880362287d70 R08: 0000000000000000 R09: 0000000000001000 [186736.681319] R10: ffff880000000000 R11: 0000000000000476 R12: 0000000000001000 [186736.681319] R13: ffff8802f91d4c88 R14: ffff8801f2a77830 R15: ffff880352e83e40 [186736.681319] FS: 00007f27b37fe700(0000) GS:ffff88043dda0000(0000) knlGS:0000000000000000 [186736.681319] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [186736.681319] CR2: 00007f27a406a000 CR3: 0000000217421000 CR4: 00000000001406e0 [186736.681319] Stack: [186736.681319] ffff880362287ea0 ffffffffa048d0bd 000000000009f000 0000000000001000 [186736.681319] 0100000000000000 ffff8801f2a77850 ffff8802f91d49b0 ffff880132600400 [186736.681319] 00000000000004f8 ffff8801c1efbe41 0000000000000000 0000000000000038 [186736.681319] Call Trace: [186736.681319] [] btrfs_ioctl+0x24cb/0x2731 [btrfs] [186736.681319] [] ? arch_local_irq_save+0x9/0xc [186736.681319] [] ? rcu_read_unlock+0x3e/0x5d [186736.681319] [] do_vfs_ioctl+0x42b/0x4ea [186736.681319] [] ? __fget_light+0x62/0x71 [186736.681319] [] SyS_ioctl+0x57/0x79 [186736.681319] [] entry_SYSCALL_64_fastpath+0x12/0x6f [186736.681319] Code: 0a 3c 6e 74 0d 3c 79 74 04 3c 59 75 0c c6 06 01 eb 03 c6 06 00 31 c0 eb 05 b8 ea ff ff ff 5d c3 55 31 c9 48 89 e5 48 39 d1 74 13 <0f> b6 04 0f 44 0f b6 04 0e 48 ff c1 44 29 c0 74 ea eb 02 31 c0 (gdb) list *(btrfs_ioctl+0x24cb) 0x5e0e1 is in btrfs_ioctl (fs/btrfs/ioctl.c:2972). 2967 dst_addr = kmap_atomic(dst_page); 2968 2969 flush_dcache_page(src_page); 2970 flush_dcache_page(dst_page); 2971 2972 if (memcmp(addr, dst_addr, cmp_len)) 2973 ret = BTRFS_SAME_DATA_DIFFERS; 2974 2975 kunmap_atomic(addr); 2976 kunmap_atomic(dst_addr); So fix this by making sure we keep the pages locked and respect the same locking order as everywhere else: get and lock the pages first and then lock the range in the inode's io tree (like for example at __btrfs_buffered_write() and extent_readpages()). If an ordered extent is found after locking the range in the io tree, unlock the range, unlock the pages, wait for the ordered extent to complete and repeat the entire locking process until no overlapping ordered extents are found. Cc: stable@vger.kernel.org # 4.2+ Signed-off-by: Filipe Manana --- fs/btrfs/ioctl.c | 90 ++++++++++++++++++++++++++++++++++++++++-------- 1 file changed, 76 insertions(+), 14 deletions(-) diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c index 83c9ad3f2621..1d6767c4c092 100644 --- a/fs/btrfs/ioctl.c +++ b/fs/btrfs/ioctl.c @@ -2811,7 +2811,6 @@ static struct page *extent_same_get_page(struct inode *inode, pgoff_t index) return NULL; } } - unlock_page(page); return page; } @@ -2830,10 +2829,17 @@ static int gather_extent_pages(struct inode *inode, struct page **pages, return 0; } -static inline void lock_extent_range(struct inode *inode, u64 off, u64 len) +static int lock_extent_range(struct inode *inode, u64 off, u64 len, + bool retry_range_locking) { - /* do any pending delalloc/csum calc on src, one way or - another, and lock file content */ + /* + * Do any pending delalloc/csum calculations on inode, one way or + * another, and lock file content. + * The locking order is: + * + * 1) pages + * 2) range in the inode's io tree + */ while (1) { struct btrfs_ordered_extent *ordered; lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1); @@ -2851,8 +2857,11 @@ static inline void lock_extent_range(struct inode *inode, u64 off, u64 len) unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1); if (ordered) btrfs_put_ordered_extent(ordered); + if (!retry_range_locking) + return -EAGAIN; btrfs_wait_ordered_range(inode, off, len); } + return 0; } static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2) @@ -2877,15 +2886,24 @@ static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1, unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1); } -static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1, - struct inode *inode2, u64 loff2, u64 len) +static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1, + struct inode *inode2, u64 loff2, u64 len, + bool retry_range_locking) { + int ret; + if (inode1 < inode2) { swap(inode1, inode2); swap(loff1, loff2); } - lock_extent_range(inode1, loff1, len); - lock_extent_range(inode2, loff2, len); + ret = lock_extent_range(inode1, loff1, len, retry_range_locking); + if (ret) + return ret; + ret = lock_extent_range(inode2, loff2, len, retry_range_locking); + if (ret) + unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, + loff1 + len - 1); + return ret; } struct cmp_pages { @@ -2901,11 +2919,15 @@ static void btrfs_cmp_data_free(struct cmp_pages *cmp) for (i = 0; i < cmp->num_pages; i++) { pg = cmp->src_pages[i]; - if (pg) + if (pg) { + unlock_page(pg); page_cache_release(pg); + } pg = cmp->dst_pages[i]; - if (pg) + if (pg) { + unlock_page(pg); page_cache_release(pg); + } } kfree(cmp->src_pages); kfree(cmp->dst_pages); @@ -2966,6 +2988,8 @@ static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst, src_page = cmp->src_pages[i]; dst_page = cmp->dst_pages[i]; + ASSERT(PageLocked(src_page)); + ASSERT(PageLocked(dst_page)); addr = kmap_atomic(src_page); dst_addr = kmap_atomic(dst_page); @@ -3078,14 +3102,46 @@ static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen, goto out_unlock; } +again: ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp); if (ret) goto out_unlock; if (same_inode) - lock_extent_range(src, same_lock_start, same_lock_len); + ret = lock_extent_range(src, same_lock_start, same_lock_len, + false); else - btrfs_double_extent_lock(src, loff, dst, dst_loff, len); + ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len, + false); + /* + * If one of the inodes has dirty pages in the respective range or + * ordered extents, we need to flush dellaloc and wait for all ordered + * extents in the range. We must unlock the pages and the ranges in the + * io trees to avoid deadlocks when flushing delalloc (requires locking + * pages) and when waiting for ordered extents to complete (they require + * range locking). + */ + if (ret == -EAGAIN) { + /* + * Ranges in the io trees already unlocked. Now unlock all + * pages before waiting for all IO to complete. + */ + btrfs_cmp_data_free(&cmp); + if (same_inode) { + btrfs_wait_ordered_range(src, same_lock_start, + same_lock_len); + } else { + btrfs_wait_ordered_range(src, loff, len); + btrfs_wait_ordered_range(dst, dst_loff, len); + } + goto again; + } + ASSERT(ret == 0); + if (WARN_ON(ret)) { + /* ranges in the io trees already unlocked */ + btrfs_cmp_data_free(&cmp); + return ret; + } /* pass original length for comparison so we stay within i_size */ ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp); @@ -3907,9 +3963,15 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, u64 lock_start = min_t(u64, off, destoff); u64 lock_len = max_t(u64, off, destoff) + len - lock_start; - lock_extent_range(src, lock_start, lock_len); + ret = lock_extent_range(src, lock_start, lock_len, true); } else { - btrfs_double_extent_lock(src, off, inode, destoff, len); + ret = btrfs_double_extent_lock(src, off, inode, destoff, len, + true); + } + ASSERT(ret == 0); + if (WARN_ON(ret)) { + /* ranges in the io trees already unlocked */ + goto out_unlock; } ret = btrfs_clone(src, inode, off, olen, len, destoff, 0); From 313140023026ae542ad76e7e268c56a1eaa2c28e Mon Sep 17 00:00:00 2001 From: Filipe Manana Date: Wed, 27 Jan 2016 18:37:47 +0000 Subject: [PATCH 2/4] Btrfs: fix page reading in extent_same ioctl leading to csum errors In the extent_same ioctl, we were grabbing the pages (locked) and attempting to read them without bothering about any concurrent IO against them. That is, we were not checking for any ongoing ordered extents nor waiting for them to complete, which leads to a race where the extent_same() code gets a checksum verification error when it reads the pages, producing a message like the following in dmesg and making the operation fail to user space with -ENOMEM: [18990.161265] BTRFS warning (device sdc): csum failed ino 259 off 495616 csum 685204116 expected csum 1515870868 Fix this by using btrfs_readpage() for reading the pages instead of extent_read_full_page_nolock(), which waits for any concurrent ordered extents to complete and locks the io range. Also do better error handling and don't treat all failures as -ENOMEM, as that's clearly misleasing, becoming identical to the checks and operation of prepare_uptodate_page(). The use of extent_read_full_page_nolock() was required before commit f441460202cb ("btrfs: fix deadlock with extent-same and readpage"), as we had the range locked in an inode's io tree before attempting to read the pages. Fixes: f441460202cb ("btrfs: fix deadlock with extent-same and readpage") Cc: stable@vger.kernel.org # 4.2+ Signed-off-by: Filipe Manana --- fs/btrfs/ioctl.c | 29 +++++++++++++++++++++-------- 1 file changed, 21 insertions(+), 8 deletions(-) diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c index 1d6767c4c092..561aa6292c1e 100644 --- a/fs/btrfs/ioctl.c +++ b/fs/btrfs/ioctl.c @@ -2794,21 +2794,27 @@ out: static struct page *extent_same_get_page(struct inode *inode, pgoff_t index) { struct page *page; - struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; page = grab_cache_page(inode->i_mapping, index); if (!page) - return NULL; + return ERR_PTR(-ENOMEM); if (!PageUptodate(page)) { - if (extent_read_full_page_nolock(tree, page, btrfs_get_extent, - 0)) - return NULL; + int ret; + + ret = btrfs_readpage(NULL, page); + if (ret) + return ERR_PTR(ret); lock_page(page); if (!PageUptodate(page)) { unlock_page(page); page_cache_release(page); - return NULL; + return ERR_PTR(-EIO); + } + if (page->mapping != inode->i_mapping) { + unlock_page(page); + page_cache_release(page); + return ERR_PTR(-EAGAIN); } } @@ -2822,9 +2828,16 @@ static int gather_extent_pages(struct inode *inode, struct page **pages, pgoff_t index = off >> PAGE_CACHE_SHIFT; for (i = 0; i < num_pages; i++) { +again: pages[i] = extent_same_get_page(inode, index + i); - if (!pages[i]) - return -ENOMEM; + if (IS_ERR(pages[i])) { + int err = PTR_ERR(pages[i]); + + if (err == -EAGAIN) + goto again; + pages[i] = NULL; + return err; + } } return 0; } From 7f042a8370a5bb7e29a6a6372e8180a56d44aa5c Mon Sep 17 00:00:00 2001 From: Filipe Manana Date: Wed, 27 Jan 2016 19:17:20 +0000 Subject: [PATCH 3/4] Btrfs: remove no longer used function extent_read_full_page_nolock() Not needed after the previous patch named "Btrfs: fix page reading in extent_same ioctl leading to csum errors". Signed-off-by: Filipe Manana --- fs/btrfs/compression.c | 6 +----- fs/btrfs/extent_io.c | 45 +++++++++++------------------------------- fs/btrfs/extent_io.h | 3 --- 3 files changed, 12 insertions(+), 42 deletions(-) diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c index c473c42d7d6c..3346cd8f9910 100644 --- a/fs/btrfs/compression.c +++ b/fs/btrfs/compression.c @@ -637,11 +637,7 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, faili = nr_pages - 1; cb->nr_pages = nr_pages; - /* In the parent-locked case, we only locked the range we are - * interested in. In all other cases, we can opportunistically - * cache decompressed data that goes beyond the requested range. */ - if (!(bio_flags & EXTENT_BIO_PARENT_LOCKED)) - add_ra_bio_pages(inode, em_start + em_len, cb); + add_ra_bio_pages(inode, em_start + em_len, cb); /* include any pages we added in add_ra-bio_pages */ uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 2e7c97a3f344..392592dc7010 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -2897,12 +2897,11 @@ static int __do_readpage(struct extent_io_tree *tree, struct block_device *bdev; int ret; int nr = 0; - int parent_locked = *bio_flags & EXTENT_BIO_PARENT_LOCKED; size_t pg_offset = 0; size_t iosize; size_t disk_io_size; size_t blocksize = inode->i_sb->s_blocksize; - unsigned long this_bio_flag = *bio_flags & EXTENT_BIO_PARENT_LOCKED; + unsigned long this_bio_flag = 0; set_page_extent_mapped(page); @@ -2942,18 +2941,16 @@ static int __do_readpage(struct extent_io_tree *tree, kunmap_atomic(userpage); set_extent_uptodate(tree, cur, cur + iosize - 1, &cached, GFP_NOFS); - if (!parent_locked) - unlock_extent_cached(tree, cur, - cur + iosize - 1, - &cached, GFP_NOFS); + unlock_extent_cached(tree, cur, + cur + iosize - 1, + &cached, GFP_NOFS); break; } em = __get_extent_map(inode, page, pg_offset, cur, end - cur + 1, get_extent, em_cached); if (IS_ERR_OR_NULL(em)) { SetPageError(page); - if (!parent_locked) - unlock_extent(tree, cur, end); + unlock_extent(tree, cur, end); break; } extent_offset = cur - em->start; @@ -3038,12 +3035,9 @@ static int __do_readpage(struct extent_io_tree *tree, set_extent_uptodate(tree, cur, cur + iosize - 1, &cached, GFP_NOFS); - if (parent_locked) - free_extent_state(cached); - else - unlock_extent_cached(tree, cur, - cur + iosize - 1, - &cached, GFP_NOFS); + unlock_extent_cached(tree, cur, + cur + iosize - 1, + &cached, GFP_NOFS); cur = cur + iosize; pg_offset += iosize; continue; @@ -3052,8 +3046,7 @@ static int __do_readpage(struct extent_io_tree *tree, if (test_range_bit(tree, cur, cur_end, EXTENT_UPTODATE, 1, NULL)) { check_page_uptodate(tree, page); - if (!parent_locked) - unlock_extent(tree, cur, cur + iosize - 1); + unlock_extent(tree, cur, cur + iosize - 1); cur = cur + iosize; pg_offset += iosize; continue; @@ -3063,8 +3056,7 @@ static int __do_readpage(struct extent_io_tree *tree, */ if (block_start == EXTENT_MAP_INLINE) { SetPageError(page); - if (!parent_locked) - unlock_extent(tree, cur, cur + iosize - 1); + unlock_extent(tree, cur, cur + iosize - 1); cur = cur + iosize; pg_offset += iosize; continue; @@ -3083,8 +3075,7 @@ static int __do_readpage(struct extent_io_tree *tree, *bio_flags = this_bio_flag; } else { SetPageError(page); - if (!parent_locked) - unlock_extent(tree, cur, cur + iosize - 1); + unlock_extent(tree, cur, cur + iosize - 1); } cur = cur + iosize; pg_offset += iosize; @@ -3213,20 +3204,6 @@ int extent_read_full_page(struct extent_io_tree *tree, struct page *page, return ret; } -int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page, - get_extent_t *get_extent, int mirror_num) -{ - struct bio *bio = NULL; - unsigned long bio_flags = EXTENT_BIO_PARENT_LOCKED; - int ret; - - ret = __do_readpage(tree, page, get_extent, NULL, &bio, mirror_num, - &bio_flags, READ, NULL); - if (bio) - ret = submit_one_bio(READ, bio, mirror_num, bio_flags); - return ret; -} - static noinline void update_nr_written(struct page *page, struct writeback_control *wbc, unsigned long nr_written) diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h index 0377413bd4b9..880d5292e972 100644 --- a/fs/btrfs/extent_io.h +++ b/fs/btrfs/extent_io.h @@ -29,7 +29,6 @@ */ #define EXTENT_BIO_COMPRESSED 1 #define EXTENT_BIO_TREE_LOG 2 -#define EXTENT_BIO_PARENT_LOCKED 4 #define EXTENT_BIO_FLAG_SHIFT 16 /* these are bit numbers for test/set bit */ @@ -210,8 +209,6 @@ static inline int lock_extent(struct extent_io_tree *tree, u64 start, u64 end) int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end); int extent_read_full_page(struct extent_io_tree *tree, struct page *page, get_extent_t *get_extent, int mirror_num); -int extent_read_full_page_nolock(struct extent_io_tree *tree, struct page *page, - get_extent_t *get_extent, int mirror_num); int __init extent_io_init(void); void extent_io_exit(void); From 0c0fe3b0fa45082cd752553fdb3a4b42503a118e Mon Sep 17 00:00:00 2001 From: Filipe Manana Date: Wed, 3 Feb 2016 19:17:27 +0000 Subject: [PATCH 4/4] Btrfs: fix hang on extent buffer lock caused by the inode_paths ioctl While doing some tests I ran into an hang on an extent buffer's rwlock that produced the following trace: [39389.800012] NMI watchdog: BUG: soft lockup - CPU#15 stuck for 22s! [fdm-stress:32166] [39389.800016] NMI watchdog: BUG: soft lockup - CPU#14 stuck for 22s! [fdm-stress:32165] [39389.800016] Modules linked in: btrfs dm_mod ppdev xor sha256_generic hmac raid6_pq drbg ansi_cprng aesni_intel i2c_piix4 acpi_cpufreq aes_x86_64 ablk_helper tpm_tis parport_pc i2c_core sg cryptd evdev psmouse lrw tpm parport gf128mul serio_raw pcspkr glue_helper processor button loop autofs4 ext4 crc16 mbcache jbd2 sd_mod sr_mod cdrom ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring crc32c_intel scsi_mod e1000 virtio floppy [last unloaded: btrfs] [39389.800016] irq event stamp: 0 [39389.800016] hardirqs last enabled at (0): [< (null)>] (null) [39389.800016] hardirqs last disabled at (0): [] copy_process+0x638/0x1a35 [39389.800016] softirqs last enabled at (0): [] copy_process+0x638/0x1a35 [39389.800016] softirqs last disabled at (0): [< (null)>] (null) [39389.800016] CPU: 14 PID: 32165 Comm: fdm-stress Not tainted 4.4.0-rc6-btrfs-next-18+ #1 [39389.800016] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS by qemu-project.org 04/01/2014 [39389.800016] task: ffff880175b1ca40 ti: ffff8800a185c000 task.ti: ffff8800a185c000 [39389.800016] RIP: 0010:[] [] queued_spin_lock_slowpath+0x57/0x158 [39389.800016] RSP: 0018:ffff8800a185fb80 EFLAGS: 00000202 [39389.800016] RAX: 0000000000000101 RBX: ffff8801710c4e9c RCX: 0000000000000101 [39389.800016] RDX: 0000000000000100 RSI: 0000000000000001 RDI: 0000000000000001 [39389.800016] RBP: ffff8800a185fb98 R08: 0000000000000001 R09: 0000000000000000 [39389.800016] R10: ffff8800a185fb68 R11: 6db6db6db6db6db7 R12: ffff8801710c4e98 [39389.800016] R13: ffff880175b1ca40 R14: ffff8800a185fc10 R15: ffff880175b1ca40 [39389.800016] FS: 00007f6d37fff700(0000) GS:ffff8802be9c0000(0000) knlGS:0000000000000000 [39389.800016] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [39389.800016] CR2: 00007f6d300019b8 CR3: 0000000037c93000 CR4: 00000000001406e0 [39389.800016] Stack: [39389.800016] ffff8801710c4e98 ffff8801710c4e98 ffff880175b1ca40 ffff8800a185fbb0 [39389.800016] ffffffff81091e11 ffff8801710c4e98 ffff8800a185fbc8 ffffffff81091895 [39389.800016] ffff8801710c4e98 ffff8800a185fbe8 ffffffff81486c5c ffffffffa067288c [39389.800016] Call Trace: [39389.800016] [] queued_read_lock_slowpath+0x46/0x60 [39389.800016] [] do_raw_read_lock+0x3e/0x41 [39389.800016] [] _raw_read_lock+0x3d/0x44 [39389.800016] [] ? btrfs_tree_read_lock+0x54/0x125 [btrfs] [39389.800016] [] btrfs_tree_read_lock+0x54/0x125 [btrfs] [39389.800016] [] ? btrfs_find_item+0xa7/0xd2 [btrfs] [39389.800016] [] btrfs_ref_to_path+0xd6/0x174 [btrfs] [39389.800016] [] inode_to_path+0x53/0xa2 [btrfs] [39389.800016] [] paths_from_inode+0x117/0x2ec [btrfs] [39389.800016] [] btrfs_ioctl+0xd5b/0x2793 [btrfs] [39389.800016] [] ? arch_local_irq_save+0x9/0xc [39389.800016] [] ? __this_cpu_preempt_check+0x13/0x15 [39389.800016] [] ? arch_local_irq_save+0x9/0xc [39389.800016] [] ? rcu_read_unlock+0x3e/0x5d [39389.800016] [] do_vfs_ioctl+0x42b/0x4ea [39389.800016] [] ? __fget_light+0x62/0x71 [39389.800016] [] SyS_ioctl+0x57/0x79 [39389.800016] [] entry_SYSCALL_64_fastpath+0x12/0x6f [39389.800016] Code: b9 01 01 00 00 f7 c6 00 ff ff ff 75 32 83 fe 01 89 ca 89 f0 0f 45 d7 f0 0f b1 13 39 f0 74 04 89 c6 eb e2 ff ca 0f 84 fa 00 00 00 <8b> 03 84 c0 74 04 f3 90 eb f6 66 c7 03 01 00 e9 e6 00 00 00 e8 [39389.800012] Modules linked in: btrfs dm_mod ppdev xor sha256_generic hmac raid6_pq drbg ansi_cprng aesni_intel i2c_piix4 acpi_cpufreq aes_x86_64 ablk_helper tpm_tis parport_pc i2c_core sg cryptd evdev psmouse lrw tpm parport gf128mul serio_raw pcspkr glue_helper processor button loop autofs4 ext4 crc16 mbcache jbd2 sd_mod sr_mod cdrom ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring crc32c_intel scsi_mod e1000 virtio floppy [last unloaded: btrfs] [39389.800012] irq event stamp: 0 [39389.800012] hardirqs last enabled at (0): [< (null)>] (null) [39389.800012] hardirqs last disabled at (0): [] copy_process+0x638/0x1a35 [39389.800012] softirqs last enabled at (0): [] copy_process+0x638/0x1a35 [39389.800012] softirqs last disabled at (0): [< (null)>] (null) [39389.800012] CPU: 15 PID: 32166 Comm: fdm-stress Tainted: G L 4.4.0-rc6-btrfs-next-18+ #1 [39389.800012] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS by qemu-project.org 04/01/2014 [39389.800012] task: ffff880179294380 ti: ffff880034a60000 task.ti: ffff880034a60000 [39389.800012] RIP: 0010:[] [] queued_write_lock_slowpath+0x62/0x72 [39389.800012] RSP: 0018:ffff880034a639f0 EFLAGS: 00000206 [39389.800012] RAX: 0000000000000101 RBX: ffff8801710c4e98 RCX: 0000000000000000 [39389.800012] RDX: 00000000000000ff RSI: 0000000000000000 RDI: ffff8801710c4e9c [39389.800012] RBP: ffff880034a639f8 R08: 0000000000000001 R09: 0000000000000000 [39389.800012] R10: ffff880034a639b0 R11: 0000000000001000 R12: ffff8801710c4e98 [39389.800012] R13: 0000000000000001 R14: ffff880172cbc000 R15: ffff8801710c4e00 [39389.800012] FS: 00007f6d377fe700(0000) GS:ffff8802be9e0000(0000) knlGS:0000000000000000 [39389.800012] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [39389.800012] CR2: 00007f6d3d3c1000 CR3: 0000000037c93000 CR4: 00000000001406e0 [39389.800012] Stack: [39389.800012] ffff8801710c4e98 ffff880034a63a10 ffffffff81091963 ffff8801710c4e98 [39389.800012] ffff880034a63a30 ffffffff81486f1b ffffffffa0672cb3 ffff8801710c4e00 [39389.800012] ffff880034a63a78 ffffffffa0672cb3 ffff8801710c4e00 ffff880034a63a58 [39389.800012] Call Trace: [39389.800012] [] do_raw_write_lock+0x72/0x8c [39389.800012] [] _raw_write_lock+0x3a/0x41 [39389.800012] [] ? btrfs_tree_lock+0x119/0x251 [btrfs] [39389.800012] [] btrfs_tree_lock+0x119/0x251 [btrfs] [39389.800012] [] ? rcu_read_unlock+0x5b/0x5d [btrfs] [39389.800012] [] ? btrfs_root_node+0xda/0xe6 [btrfs] [39389.800012] [] btrfs_lock_root_node+0x22/0x42 [btrfs] [39389.800012] [] btrfs_search_slot+0x1b8/0x758 [btrfs] [39389.800012] [] ? time_hardirqs_on+0x15/0x28 [39389.800012] [] btrfs_lookup_inode+0x31/0x95 [btrfs] [39389.800012] [] ? trace_hardirqs_on+0xd/0xf [39389.800012] [] ? mutex_lock_nested+0x397/0x3bc [39389.800012] [] __btrfs_update_delayed_inode+0x59/0x1c0 [btrfs] [39389.800012] [] __btrfs_commit_inode_delayed_items+0x194/0x5aa [btrfs] [39389.800012] [] ? _raw_spin_unlock+0x31/0x44 [39389.800012] [] __btrfs_run_delayed_items+0xa4/0x15c [btrfs] [39389.800012] [] btrfs_run_delayed_items+0x11/0x13 [btrfs] [39389.800012] [] btrfs_commit_transaction+0x234/0x96e [btrfs] [39389.800012] [] btrfs_sync_fs+0x145/0x1ad [btrfs] [39389.800012] [] btrfs_ioctl+0x11d2/0x2793 [btrfs] [39389.800012] [] ? arch_local_irq_save+0x9/0xc [39389.800012] [] ? __might_fault+0x4c/0xa7 [39389.800012] [] ? __might_fault+0x4c/0xa7 [39389.800012] [] ? arch_local_irq_save+0x9/0xc [39389.800012] [] ? rcu_read_unlock+0x3e/0x5d [39389.800012] [] do_vfs_ioctl+0x42b/0x4ea [39389.800012] [] ? __fget_light+0x62/0x71 [39389.800012] [] SyS_ioctl+0x57/0x79 [39389.800012] [] entry_SYSCALL_64_fastpath+0x12/0x6f [39389.800012] Code: f0 0f b1 13 85 c0 75 ef eb 2a f3 90 8a 03 84 c0 75 f8 f0 0f b0 13 84 c0 75 f0 ba ff 00 00 00 eb 0a f0 0f b1 13 ff c8 74 0b f3 90 <8b> 03 83 f8 01 75 f7 eb ed c6 43 04 00 5b 5d c3 0f 1f 44 00 00 This happens because in the code path executed by the inode_paths ioctl we end up nesting two calls to read lock a leaf's rwlock when after the first call to read_lock() and before the second call to read_lock(), another task (running the delayed items as part of a transaction commit) has already called write_lock() against the leaf's rwlock. This situation is illustrated by the following diagram: Task A Task B btrfs_ref_to_path() btrfs_commit_transaction() read_lock(&eb->lock); btrfs_run_delayed_items() __btrfs_commit_inode_delayed_items() __btrfs_update_delayed_inode() btrfs_lookup_inode() write_lock(&eb->lock); --> task waits for lock read_lock(&eb->lock); --> makes this task hang forever (and task B too of course) So fix this by avoiding doing the nested read lock, which is easily avoidable. This issue does not happen if task B calls write_lock() after task A does the second call to read_lock(), however there does not seem to exist anything in the documentation that mentions what is the expected behaviour for recursive locking of rwlocks (leaving the idea that doing so is not a good usage of rwlocks). Also, as a side effect necessary for this fix, make sure we do not needlessly read lock extent buffers when the input path has skip_locking set (used when called from send). Cc: stable@vger.kernel.org Signed-off-by: Filipe Manana --- fs/btrfs/backref.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) diff --git a/fs/btrfs/backref.c b/fs/btrfs/backref.c index b90cd3776f8e..f6dac40f87ff 100644 --- a/fs/btrfs/backref.c +++ b/fs/btrfs/backref.c @@ -1406,7 +1406,8 @@ char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, read_extent_buffer(eb, dest + bytes_left, name_off, name_len); if (eb != eb_in) { - btrfs_tree_read_unlock_blocking(eb); + if (!path->skip_locking) + btrfs_tree_read_unlock_blocking(eb); free_extent_buffer(eb); } ret = btrfs_find_item(fs_root, path, parent, 0, @@ -1426,9 +1427,10 @@ char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, eb = path->nodes[0]; /* make sure we can use eb after releasing the path */ if (eb != eb_in) { - atomic_inc(&eb->refs); - btrfs_tree_read_lock(eb); - btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); + if (!path->skip_locking) + btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); + path->nodes[0] = NULL; + path->locks[0] = 0; } btrfs_release_path(path); iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);