diff options
Diffstat (limited to 'sys/contrib/openzfs/module/os/linux/zfs')
8 files changed, 255 insertions, 30 deletions
diff --git a/sys/contrib/openzfs/module/os/linux/zfs/abd_os.c b/sys/contrib/openzfs/module/os/linux/zfs/abd_os.c index 8a8316f63c48..18f2426fbbfc 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/abd_os.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/abd_os.c @@ -23,6 +23,7 @@ * Copyright (c) 2014 by Chunwei Chen. All rights reserved. * Copyright (c) 2019 by Delphix. All rights reserved. * Copyright (c) 2023, 2024, Klara Inc. + * Copyright (c) 2025, Rob Norris <robn@despairlabs.com> */ /* @@ -1109,6 +1110,14 @@ abd_return_buf_copy(abd_t *abd, void *buf, size_t n) #define ABD_ITER_PAGE_SIZE(page) (PAGESIZE) #endif +#ifndef nth_page +/* + * Since 6.18 nth_page() no longer exists, and is no longer required to iterate + * within a single SG entry, so we replace it with a simple addition. + */ +#define nth_page(p, n) ((p)+(n)) +#endif + void abd_iter_page(struct abd_iter *aiter) { diff --git a/sys/contrib/openzfs/module/os/linux/zfs/vdev_disk.c b/sys/contrib/openzfs/module/os/linux/zfs/vdev_disk.c index 830fad7fe793..1bd3500e9f66 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/vdev_disk.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/vdev_disk.c @@ -471,13 +471,17 @@ vdev_disk_close(vdev_t *v) if (v->vdev_reopening || vd == NULL) return; + rw_enter(&vd->vd_lock, RW_WRITER); + if (vd->vd_bdh != NULL) vdev_blkdev_put(vd->vd_bdh, spa_mode(v->vdev_spa), zfs_vdev_holder); + v->vdev_tsd = NULL; + + rw_exit(&vd->vd_lock); rw_destroy(&vd->vd_lock); kmem_free(vd, sizeof (vdev_disk_t)); - v->vdev_tsd = NULL; } /* diff --git a/sys/contrib/openzfs/module/os/linux/zfs/zfs_acl.c b/sys/contrib/openzfs/module/os/linux/zfs/zfs_acl.c index daa4b5776837..934d74a112fd 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/zfs_acl.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/zfs_acl.c @@ -2524,7 +2524,7 @@ zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode, * Also note: DOS R/O is ignored for directories. */ if ((v4_mode & WRITE_MASK_DATA) && - S_ISDIR(ZTOI(zp)->i_mode) && + !S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_READONLY)) { return (SET_ERROR(EPERM)); } diff --git a/sys/contrib/openzfs/module/os/linux/zfs/zfs_vfsops.c b/sys/contrib/openzfs/module/os/linux/zfs/zfs_vfsops.c index cd606e667bff..8a7d14ab6119 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/zfs_vfsops.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/zfs_vfsops.c @@ -1556,6 +1556,12 @@ zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent) sb->s_xattr = zpl_xattr_handlers; sb->s_export_op = &zpl_export_operations; +#ifdef HAVE_SET_DEFAULT_D_OP + set_default_d_op(sb, &zpl_dentry_operations); +#else + sb->s_d_op = &zpl_dentry_operations; +#endif + /* Set features for file system. */ zfs_set_fuid_feature(zfsvfs); diff --git a/sys/contrib/openzfs/module/os/linux/zfs/zfs_vnops_os.c b/sys/contrib/openzfs/module/os/linux/zfs/zfs_vnops_os.c index 6106726651a3..e845ad69ad78 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/zfs_vnops_os.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/zfs_vnops_os.c @@ -2033,10 +2033,7 @@ zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr, zidmap_t *mnt_ns) goto out3; } - if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) { - err = SET_ERROR(EPERM); - goto out3; - } + /* ZFS_READONLY will be handled in zfs_zaccess() */ /* * Verify timestamps doesn't overflow 32 bits. diff --git a/sys/contrib/openzfs/module/os/linux/zfs/zpl_file.c b/sys/contrib/openzfs/module/os/linux/zfs/zpl_file.c index d07317b0d910..02965ac8cbee 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/zpl_file.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/zpl_file.c @@ -23,6 +23,7 @@ * Copyright (c) 2011, Lawrence Livermore National Security, LLC. * Copyright (c) 2015 by Chunwei Chen. All rights reserved. * Copyright (c) 2025, Klara, Inc. + * Copyright (c) 2025, Rob Norris <robn@despairlabs.com> */ @@ -478,6 +479,7 @@ zpl_putpage(struct page *pp, struct writeback_control *wbc, void *data) return (ret); } +#ifdef HAVE_WRITE_CACHE_PAGES #ifdef HAVE_WRITEPAGE_T_FOLIO static int zpl_putfolio(struct folio *pp, struct writeback_control *wbc, void *data) @@ -499,6 +501,78 @@ zpl_write_cache_pages(struct address_space *mapping, #endif return (result); } +#else +static inline int +zpl_write_cache_pages(struct address_space *mapping, + struct writeback_control *wbc, void *data) +{ + pgoff_t start = wbc->range_start >> PAGE_SHIFT; + pgoff_t end = wbc->range_end >> PAGE_SHIFT; + + struct folio_batch fbatch; + folio_batch_init(&fbatch); + + /* + * This atomically (-ish) tags all DIRTY pages in the range with + * TOWRITE, allowing users to continue dirtying or undirtying pages + * while we get on with writeback, without us treading on each other. + */ + tag_pages_for_writeback(mapping, start, end); + + int err = 0; + unsigned int npages; + + /* + * Grab references to the TOWRITE pages just flagged. This may not get + * all of them, so we do it in a loop until there are none left. + */ + while ((npages = filemap_get_folios_tag(mapping, &start, end, + PAGECACHE_TAG_TOWRITE, &fbatch)) != 0) { + + /* Loop over each page and write it out. */ + struct folio *folio; + while ((folio = folio_batch_next(&fbatch)) != NULL) { + folio_lock(folio); + + /* + * If the folio has been remapped, or is no longer + * dirty, then there's nothing to do. + */ + if (folio->mapping != mapping || + !folio_test_dirty(folio)) { + folio_unlock(folio); + continue; + } + + /* + * If writeback is already in progress, wait for it to + * finish. We continue after this even if the page + * ends up clean; zfs_putpage() will skip it if no + * further work is required. + */ + while (folio_test_writeback(folio)) + folio_wait_bit(folio, PG_writeback); + + /* + * Write it out and collect any error. zfs_putpage() + * will clear the TOWRITE and DIRTY flags, and return + * with the page unlocked. + */ + int ferr = zpl_putpage(&folio->page, wbc, data); + if (err == 0 && ferr != 0) + err = ferr; + + /* Housekeeping for the caller. */ + wbc->nr_to_write -= folio_nr_pages(folio); + } + + /* Release any remaining references on the batch. */ + folio_batch_release(&fbatch); + } + + return (err); +} +#endif static int zpl_writepages(struct address_space *mapping, struct writeback_control *wbc) diff --git a/sys/contrib/openzfs/module/os/linux/zfs/zpl_super.c b/sys/contrib/openzfs/module/os/linux/zfs/zpl_super.c index 53819628627d..347b352506e5 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/zpl_super.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/zpl_super.c @@ -22,6 +22,8 @@ /* * Copyright (c) 2011, Lawrence Livermore National Security, LLC. * Copyright (c) 2023, Datto Inc. All rights reserved. + * Copyright (c) 2025, Klara, Inc. + * Copyright (c) 2025, Rob Norris <robn@despairlabs.com> */ @@ -32,7 +34,22 @@ #include <sys/zpl.h> #include <linux/iversion.h> #include <linux/version.h> +#include <linux/vfs_compat.h> +/* + * What to do when the last reference to an inode is released. If 0, the kernel + * will cache it on the superblock. If 1, the inode will be freed immediately. + * See zpl_drop_inode(). + */ +int zfs_delete_inode = 0; + +/* + * What to do when the last reference to a dentry is released. If 0, the kernel + * will cache it until the entry (file) is destroyed. If 1, the dentry will be + * marked for cleanup, at which time its inode reference will be released. See + * zpl_dentry_delete(). + */ +int zfs_delete_dentry = 0; static struct inode * zpl_inode_alloc(struct super_block *sb) @@ -77,11 +94,36 @@ zpl_dirty_inode(struct inode *ip, int flags) } /* - * When ->drop_inode() is called its return value indicates if the - * inode should be evicted from the inode cache. If the inode is - * unhashed and has no links the default policy is to evict it - * immediately. + * ->drop_inode() is called when the last reference to an inode is released. + * Its return value indicates if the inode should be destroyed immediately, or + * cached on the superblock structure. + * + * By default (zfs_delete_inode=0), we call generic_drop_inode(), which returns + * "destroy immediately" if the inode is unhashed and has no links (roughly: no + * longer exists on disk). On datasets with millions of rarely-accessed files, + * this can cause a large amount of memory to be "pinned" by cached inodes, + * which in turn pin their associated dnodes and dbufs, until the kernel starts + * reporting memory pressure and requests OpenZFS release some memory (see + * zfs_prune()). + * + * When set to 1, we call generic_delete_inode(), which always returns "destroy + * immediately", resulting in inodes being destroyed immediately, releasing + * their associated dnodes and dbufs to the dbuf cached and the ARC to be + * evicted as normal. * + * Note that the "last reference" doesn't always mean the last _userspace_ + * reference; the dentry cache also holds a reference, so "busy" inodes will + * still be kept alive that way (subject to dcache tuning). + */ +static int +zpl_drop_inode(struct inode *ip) +{ + if (zfs_delete_inode) + return (generic_delete_inode(ip)); + return (generic_drop_inode(ip)); +} + +/* * The ->evict_inode() callback must minimally truncate the inode pages, * and call clear_inode(). For 2.6.35 and later kernels this will * simply update the inode state, with the sync occurring before the @@ -470,6 +512,7 @@ const struct super_operations zpl_super_operations = { .destroy_inode = zpl_inode_destroy, .dirty_inode = zpl_dirty_inode, .write_inode = NULL, + .drop_inode = zpl_drop_inode, .evict_inode = zpl_evict_inode, .put_super = zpl_put_super, .sync_fs = zpl_sync_fs, @@ -480,6 +523,35 @@ const struct super_operations zpl_super_operations = { .show_stats = NULL, }; +/* + * ->d_delete() is called when the last reference to a dentry is released. Its + * return value indicates if the dentry should be destroyed immediately, or + * retained in the dentry cache. + * + * By default (zfs_delete_dentry=0) the kernel will always cache unused + * entries. Each dentry holds an inode reference, so cached dentries can hold + * the final inode reference indefinitely, leading to the inode and its related + * data being pinned (see zpl_drop_inode()). + * + * When set to 1, we signal that the dentry should be destroyed immediately and + * never cached. This reduces memory usage, at the cost of higher overheads to + * lookup a file, as the inode and its underlying data (dnode/dbuf) need to be + * reloaded and reinflated. + * + * Note that userspace does not have direct control over dentry references and + * reclaim; rather, this is part of the kernel's caching and reclaim subsystems + * (eg vm.vfs_cache_pressure). + */ +static int +zpl_dentry_delete(const struct dentry *dentry) +{ + return (zfs_delete_dentry ? 1 : 0); +} + +const struct dentry_operations zpl_dentry_operations = { + .d_delete = zpl_dentry_delete, +}; + struct file_system_type zpl_fs_type = { .owner = THIS_MODULE, .name = ZFS_DRIVER, @@ -491,3 +563,10 @@ struct file_system_type zpl_fs_type = { .mount = zpl_mount, .kill_sb = zpl_kill_sb, }; + +ZFS_MODULE_PARAM(zfs, zfs_, delete_inode, INT, ZMOD_RW, + "Delete inodes as soon as the last reference is released."); + +ZFS_MODULE_PARAM(zfs, zfs_, delete_dentry, INT, ZMOD_RW, + "Delete dentries from dentry cache as soon as the last reference is " + "released."); diff --git a/sys/contrib/openzfs/module/os/linux/zfs/zvol_os.c b/sys/contrib/openzfs/module/os/linux/zfs/zvol_os.c index bac166fcd89e..fe939150b641 100644 --- a/sys/contrib/openzfs/module/os/linux/zfs/zvol_os.c +++ b/sys/contrib/openzfs/module/os/linux/zfs/zvol_os.c @@ -21,7 +21,7 @@ */ /* * Copyright (c) 2012, 2020 by Delphix. All rights reserved. - * Copyright (c) 2024, Rob Norris <robn@despairlabs.com> + * Copyright (c) 2024, 2025, Rob Norris <robn@despairlabs.com> * Copyright (c) 2024, 2025, Klara, Inc. */ @@ -337,16 +337,14 @@ zvol_discard(zv_request_t *zvr) } /* - * Align the request to volume block boundaries when a secure erase is - * not required. This will prevent dnode_free_range() from zeroing out - * the unaligned parts which is slow (read-modify-write) and useless - * since we are not freeing any space by doing so. + * Align the request to volume block boundaries. This will prevent + * dnode_free_range() from zeroing out the unaligned parts which is + * slow (read-modify-write) and useless since we are not freeing any + * space by doing so. */ - if (!io_is_secure_erase(bio, rq)) { - start = P2ROUNDUP(start, zv->zv_volblocksize); - end = P2ALIGN_TYPED(end, zv->zv_volblocksize, uint64_t); - size = end - start; - } + start = P2ROUNDUP(start, zv->zv_volblocksize); + end = P2ALIGN_TYPED(end, zv->zv_volblocksize, uint64_t); + size = end - start; if (start >= end) goto unlock; @@ -467,6 +465,24 @@ zvol_read_task(void *arg) zv_request_task_free(task); } +/* + * Note: + * + * The kernel uses different enum names for the IO opcode, depending on the + * kernel version ('req_opf', 'req_op'). To sidestep this, use macros rather + * than inline functions for these checks. + */ +/* Should this IO go down the zvol write path? */ +#define ZVOL_OP_IS_WRITE(op) \ + (op == REQ_OP_WRITE || \ + op == REQ_OP_FLUSH || \ + op == REQ_OP_DISCARD) + +/* Is this IO type supported by zvols? */ +#define ZVOL_OP_IS_SUPPORTED(op) (op == REQ_OP_READ || ZVOL_OP_IS_WRITE(op)) + +/* Get the IO opcode */ +#define ZVOL_OP(bio, rq) (bio != NULL ? bio_op(bio) : req_op(rq)) /* * Process a BIO or request @@ -484,7 +500,33 @@ zvol_request_impl(zvol_state_t *zv, struct bio *bio, struct request *rq, fstrans_cookie_t cookie = spl_fstrans_mark(); uint64_t offset = io_offset(bio, rq); uint64_t size = io_size(bio, rq); - int rw = io_data_dir(bio, rq); + int rw; + + if (unlikely(!ZVOL_OP_IS_SUPPORTED(ZVOL_OP(bio, rq)))) { + zfs_dbgmsg("Unsupported zvol %s, op=%d, flags=0x%x", + rq != NULL ? "request" : "BIO", + ZVOL_OP(bio, rq), + rq != NULL ? rq->cmd_flags : bio->bi_opf); + ASSERT(ZVOL_OP_IS_SUPPORTED(ZVOL_OP(bio, rq))); + zvol_end_io(bio, rq, SET_ERROR(ENOTSUPP)); + goto out; + } + + if (ZVOL_OP_IS_WRITE(ZVOL_OP(bio, rq))) { + rw = WRITE; + } else { + rw = READ; + } + + /* + * Sanity check + * + * If we're a BIO, check our rw matches the kernel's + * bio_data_dir(bio) rw. We need to check because we support fewer + * IO operations, and want to verify that what we think are reads and + * writes from those operations match what the kernel thinks. + */ + ASSERT(rq != NULL || rw == bio_data_dir(bio)); if (unlikely(zv->zv_flags & ZVOL_REMOVING)) { zvol_end_io(bio, rq, SET_ERROR(ENXIO)); @@ -589,7 +631,7 @@ zvol_request_impl(zvol_state_t *zv, struct bio *bio, struct request *rq, * interfaces lack this functionality (they block waiting for * the i/o to complete). */ - if (io_is_discard(bio, rq) || io_is_secure_erase(bio, rq)) { + if (io_is_discard(bio, rq)) { if (force_sync) { zvol_discard(&zvr); } else { @@ -990,12 +1032,12 @@ zvol_os_update_volsize(zvol_state_t *zv, uint64_t volsize) * tiny devices. For devices over 1 Mib a standard head and sector count * is used to keep the cylinders count reasonable. */ -static int -zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo) +static inline int +zvol_getgeo_impl(struct gendisk *disk, struct hd_geometry *geo) { + zvol_state_t *zv = atomic_load_ptr(&disk->private_data); sector_t sectors; - zvol_state_t *zv = atomic_load_ptr(&bdev->bd_disk->private_data); ASSERT3P(zv, !=, NULL); ASSERT3U(zv->zv_open_count, >, 0); @@ -1015,6 +1057,20 @@ zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo) return (0); } +#ifdef HAVE_BLOCK_DEVICE_OPERATIONS_GETGEO_GENDISK +static int +zvol_getgeo(struct gendisk *disk, struct hd_geometry *geo) +{ + return (zvol_getgeo_impl(disk, geo)); +} +#else +static int +zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo) +{ + return (zvol_getgeo_impl(bdev->bd_disk, geo)); +} +#endif + /* * Why have two separate block_device_operations structs? * @@ -1458,7 +1514,7 @@ zvol_os_remove_minor(zvol_state_t *zv) if (zso->use_blk_mq) blk_mq_free_tag_set(&zso->tag_set); - ida_simple_remove(&zvol_ida, MINOR(zso->zvo_dev) >> ZVOL_MINOR_BITS); + ida_free(&zvol_ida, MINOR(zso->zvo_dev) >> ZVOL_MINOR_BITS); kmem_free(zso, sizeof (struct zvol_state_os)); @@ -1613,7 +1669,7 @@ zvol_os_create_minor(const char *name) if (zvol_inhibit_dev) return (0); - idx = ida_simple_get(&zvol_ida, 0, 0, kmem_flags_convert(KM_SLEEP)); + idx = ida_alloc(&zvol_ida, kmem_flags_convert(KM_SLEEP)); if (idx < 0) return (SET_ERROR(-idx)); minor = idx << ZVOL_MINOR_BITS; @@ -1621,7 +1677,7 @@ zvol_os_create_minor(const char *name) /* too many partitions can cause an overflow */ zfs_dbgmsg("zvol: create minor overflow: %s, minor %u/%u", name, minor, MINOR(minor)); - ida_simple_remove(&zvol_ida, idx); + ida_free(&zvol_ida, idx); return (SET_ERROR(EINVAL)); } @@ -1629,7 +1685,7 @@ zvol_os_create_minor(const char *name) if (zv) { ASSERT(MUTEX_HELD(&zv->zv_state_lock)); mutex_exit(&zv->zv_state_lock); - ida_simple_remove(&zvol_ida, idx); + ida_free(&zvol_ida, idx); return (SET_ERROR(EEXIST)); } @@ -1729,7 +1785,7 @@ out_doi: rw_exit(&zvol_state_lock); error = zvol_os_add_disk(zv->zv_zso->zvo_disk); } else { - ida_simple_remove(&zvol_ida, idx); + ida_free(&zvol_ida, idx); } return (error); |