diff options
Diffstat (limited to 'fs/btrfs/send.c')
| -rw-r--r-- | fs/btrfs/send.c | 5352 |
1 files changed, 3397 insertions, 1955 deletions
diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c index 7ea2d6b1f170..2522faa97478 100644 --- a/fs/btrfs/send.c +++ b/fs/btrfs/send.c @@ -4,6 +4,7 @@ */ #include <linux/bsearch.h> +#include <linux/falloc.h> #include <linux/fs.h> #include <linux/file.h> #include <linux/sort.h> @@ -15,14 +16,30 @@ #include <linux/string.h> #include <linux/compat.h> #include <linux/crc32c.h> - +#include <linux/fsverity.h> #include "send.h" +#include "ctree.h" #include "backref.h" #include "locking.h" #include "disk-io.h" #include "btrfs_inode.h" #include "transaction.h" #include "compression.h" +#include "print-tree.h" +#include "accessors.h" +#include "dir-item.h" +#include "file-item.h" +#include "ioctl.h" +#include "verity.h" +#include "lru_cache.h" + +/* + * Maximum number of references an extent can have in order for us to attempt to + * issue clone operations instead of write operations. This currently exists to + * avoid hitting limitations of the backreference walking code (taking a lot of + * time and using too much memory for extents with large number of references). + */ +#define SEND_MAX_EXTENT_REFS 1024 /* * A fs_path is a helper to dynamically build path names with unknown size. @@ -30,28 +47,30 @@ * It allows fast adding of path elements on the right side (normal path) and * fast adding to the left side (reversed path). A reversed path can also be * unreversed if needed. + * + * The definition of struct fs_path relies on -fms-extensions to allow + * including a tagged struct as an anonymous member. */ +struct __fs_path { + char *start; + char *end; + + char *buf; + unsigned short buf_len:15; + unsigned short reversed:1; +}; +static_assert(sizeof(struct __fs_path) < 256); struct fs_path { - union { - struct { - char *start; - char *end; - - char *buf; - unsigned short buf_len:15; - unsigned short reversed:1; - char inline_buf[]; - }; - /* - * Average path length does not exceed 200 bytes, we'll have - * better packing in the slab and higher chance to satisfy - * a allocation later during send. - */ - char pad[256]; - }; + struct __fs_path; + /* + * Average path length does not exceed 200 bytes, we'll have + * better packing in the slab and higher chance to satisfy + * an allocation later during send. + */ + char inline_buf[256 - sizeof(struct __fs_path)]; }; #define FS_PATH_INLINE_SIZE \ - (sizeof(struct fs_path) - offsetof(struct fs_path, inline_buf)) + sizeof_field(struct fs_path, inline_buf) /* reused for each extent */ @@ -59,12 +78,61 @@ struct clone_root { struct btrfs_root *root; u64 ino; u64 offset; + u64 num_bytes; + bool found_ref; +}; + +#define SEND_MAX_NAME_CACHE_SIZE 256 + +/* + * Limit the root_ids array of struct backref_cache_entry to 17 elements. + * This makes the size of a cache entry to be exactly 192 bytes on x86_64, which + * can be satisfied from the kmalloc-192 slab, without wasting any space. + * The most common case is to have a single root for cloning, which corresponds + * to the send root. Having the user specify more than 16 clone roots is not + * common, and in such rare cases we simply don't use caching if the number of + * cloning roots that lead down to a leaf is more than 17. + */ +#define SEND_MAX_BACKREF_CACHE_ROOTS 17 - u64 found_refs; +/* + * Max number of entries in the cache. + * With SEND_MAX_BACKREF_CACHE_ROOTS as 17, the size in bytes, excluding + * maple tree's internal nodes, is 24K. + */ +#define SEND_MAX_BACKREF_CACHE_SIZE 128 + +/* + * A backref cache entry maps a leaf to a list of IDs of roots from which the + * leaf is accessible and we can use for clone operations. + * With SEND_MAX_BACKREF_CACHE_ROOTS as 12, each cache entry is 128 bytes (on + * x86_64). + */ +struct backref_cache_entry { + struct btrfs_lru_cache_entry entry; + u64 root_ids[SEND_MAX_BACKREF_CACHE_ROOTS]; + /* Number of valid elements in the root_ids array. */ + int num_roots; }; -#define SEND_CTX_MAX_NAME_CACHE_SIZE 128 -#define SEND_CTX_NAME_CACHE_CLEAN_SIZE (SEND_CTX_MAX_NAME_CACHE_SIZE * 2) +/* See the comment at lru_cache.h about struct btrfs_lru_cache_entry. */ +static_assert(offsetof(struct backref_cache_entry, entry) == 0); + +/* + * Max number of entries in the cache that stores directories that were already + * created. The cache uses raw struct btrfs_lru_cache_entry entries, so it uses + * at most 4096 bytes - sizeof(struct btrfs_lru_cache_entry) is 48 bytes, but + * the kmalloc-64 slab is used, so we get 4096 bytes (64 bytes * 64). + */ +#define SEND_MAX_DIR_CREATED_CACHE_SIZE 64 + +/* + * Max number of entries in the cache that stores directories that were already + * created. The cache uses raw struct btrfs_lru_cache_entry entries, so it uses + * at most 4096 bytes - sizeof(struct btrfs_lru_cache_entry) is 48 bytes, but + * the kmalloc-64 slab is used, so we get 4096 bytes (64 bytes * 64). + */ +#define SEND_MAX_DIR_UTIMES_CACHE_SIZE 64 struct send_ctx { struct file *send_filp; @@ -72,9 +140,15 @@ struct send_ctx { char *send_buf; u32 send_size; u32 send_max_size; - u64 total_send_size; - u64 cmd_send_size[BTRFS_SEND_C_MAX + 1]; + /* + * Whether BTRFS_SEND_A_DATA attribute was already added to current + * command (since protocol v2, data must be the last attribute). + */ + bool put_data; + struct page **send_buf_pages; u64 flags; /* 'flags' member of btrfs_ioctl_send_args is u64 */ + /* Protocol version compatibility requested */ + u32 proto; struct btrfs_root *send_root; struct btrfs_root *parent_root; @@ -87,33 +161,47 @@ struct send_ctx { struct btrfs_key *cmp_key; /* + * Keep track of the generation of the last transaction that was used + * for relocating a block group. This is periodically checked in order + * to detect if a relocation happened since the last check, so that we + * don't operate on stale extent buffers for nodes (level >= 1) or on + * stale disk_bytenr values of file extent items. + */ + u64 last_reloc_trans; + + /* * infos of the currently processed inode. In case of deleted inodes, * these are the values from the deleted inode. */ u64 cur_ino; u64 cur_inode_gen; - int cur_inode_new; - int cur_inode_new_gen; - int cur_inode_deleted; u64 cur_inode_size; u64 cur_inode_mode; u64 cur_inode_rdev; u64 cur_inode_last_extent; u64 cur_inode_next_write_offset; + bool cur_inode_new; + bool cur_inode_new_gen; + bool cur_inode_deleted; bool ignore_cur_inode; + bool cur_inode_needs_verity; + void *verity_descriptor; u64 send_progress; struct list_head new_refs; struct list_head deleted_refs; - struct radix_tree_root name_cache; - struct list_head name_cache_list; - int name_cache_size; + struct btrfs_lru_cache name_cache; + /* + * The inode we are currently processing. It's not NULL only when we + * need to issue write commands for data extents from this inode. + */ + struct inode *cur_inode; struct file_ra_state ra; - - char *read_buf; + u64 page_cache_clear_start; + bool clean_page_cache; /* * We process inodes by their increasing order, so if before an @@ -209,6 +297,17 @@ struct send_ctx { * Indexed by the inode number of the directory to be deleted. */ struct rb_root orphan_dirs; + + struct rb_root rbtree_new_refs; + struct rb_root rbtree_deleted_refs; + + struct btrfs_lru_cache backref_cache; + u64 backref_cache_last_reloc_trans; + + struct btrfs_lru_cache dir_created_cache; + struct btrfs_lru_cache dir_utimes_cache; + + struct fs_path cur_inode_path; }; struct pending_dir_move { @@ -229,6 +328,7 @@ struct waiting_dir_move { * after this directory is moved, we can try to rmdir the ino rmdir_ino. */ u64 rmdir_ino; + u64 rmdir_gen; bool orphanized; }; @@ -237,27 +337,36 @@ struct orphan_dir_info { u64 ino; u64 gen; u64 last_dir_index_offset; + u64 dir_high_seq_ino; }; struct name_cache_entry { - struct list_head list; /* - * radix_tree has only 32bit entries but we need to handle 64bit inums. - * We use the lower 32bit of the 64bit inum to store it in the tree. If - * more then one inum would fall into the same entry, we use radix_list - * to store the additional entries. radix_list is also used to store - * entries where two entries have the same inum but different - * generations. + * The key in the entry is an inode number, and the generation matches + * the inode's generation. */ - struct list_head radix_list; - u64 ino; - u64 gen; + struct btrfs_lru_cache_entry entry; u64 parent_ino; u64 parent_gen; int ret; int need_later_update; + /* Name length without NUL terminator. */ int name_len; - char name[]; + /* Not NUL terminated. */ + char name[] __counted_by(name_len) __nonstring; +}; + +/* See the comment at lru_cache.h about struct btrfs_lru_cache_entry. */ +static_assert(offsetof(struct name_cache_entry, entry) == 0); + +#define ADVANCE 1 +#define ADVANCE_ONLY_NEXT -1 + +enum btrfs_compare_tree_result { + BTRFS_COMPARE_TREE_NEW, + BTRFS_COMPARE_TREE_DELETED, + BTRFS_COMPARE_TREE_CHANGED, + BTRFS_COMPARE_TREE_SAME, }; __cold @@ -278,20 +387,30 @@ static void inconsistent_snapshot_error(struct send_ctx *sctx, result_string = "updated"; break; case BTRFS_COMPARE_TREE_SAME: - ASSERT(0); + DEBUG_WARN("no change between trees"); result_string = "unchanged"; break; default: - ASSERT(0); + DEBUG_WARN("unexpected comparison result %d", result); result_string = "unexpected"; } btrfs_err(sctx->send_root->fs_info, "Send: inconsistent snapshot, found %s %s for inode %llu without updated inode item, send root is %llu, parent root is %llu", result_string, what, sctx->cmp_key->objectid, - sctx->send_root->root_key.objectid, - (sctx->parent_root ? - sctx->parent_root->root_key.objectid : 0)); + btrfs_root_id(sctx->send_root), + (sctx->parent_root ? btrfs_root_id(sctx->parent_root) : 0)); +} + +__maybe_unused +static bool proto_cmd_ok(const struct send_ctx *sctx, int cmd) +{ + switch (sctx->proto) { + case 1: return cmd <= BTRFS_SEND_C_MAX_V1; + case 2: return cmd <= BTRFS_SEND_C_MAX_V2; + case 3: return cmd <= BTRFS_SEND_C_MAX_V3; + default: return false; + } } static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); @@ -299,7 +418,7 @@ static int is_waiting_for_move(struct send_ctx *sctx, u64 ino); static struct waiting_dir_move * get_waiting_dir_move(struct send_ctx *sctx, u64 ino); -static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino); +static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino, u64 gen); static int need_send_hole(struct send_ctx *sctx) { @@ -310,15 +429,21 @@ static int need_send_hole(struct send_ctx *sctx) static void fs_path_reset(struct fs_path *p) { - if (p->reversed) { + if (p->reversed) p->start = p->buf + p->buf_len - 1; - p->end = p->start; - *p->start = 0; - } else { + else p->start = p->buf; - p->end = p->start; - *p->start = 0; - } + + p->end = p->start; + *p->start = 0; +} + +static void init_path(struct fs_path *p) +{ + p->reversed = 0; + p->buf = p->inline_buf; + p->buf_len = FS_PATH_INLINE_SIZE; + fs_path_reset(p); } static struct fs_path *fs_path_alloc(void) @@ -328,10 +453,7 @@ static struct fs_path *fs_path_alloc(void) p = kmalloc(sizeof(*p), GFP_KERNEL); if (!p) return NULL; - p->reversed = 0; - p->buf = p->inline_buf; - p->buf_len = FS_PATH_INLINE_SIZE; - fs_path_reset(p); + init_path(p); return p; } @@ -356,7 +478,7 @@ static void fs_path_free(struct fs_path *p) kfree(p); } -static int fs_path_len(struct fs_path *p) +static inline int fs_path_len(const struct fs_path *p) { return p->end - p->start; } @@ -372,15 +494,18 @@ static int fs_path_ensure_buf(struct fs_path *p, int len) if (p->buf_len >= len) return 0; - if (len > PATH_MAX) { - WARN_ON(1); - return -ENOMEM; - } + if (WARN_ON(len > PATH_MAX)) + return -ENAMETOOLONG; - path_len = p->end - p->start; + path_len = fs_path_len(p); old_buf_len = p->buf_len; /* + * Allocate to the next largest kmalloc bucket size, to let + * the fast path happen most of the time. + */ + len = kmalloc_size_roundup(len); + /* * First time the inline_buf does not suffice */ if (p->buf == p->inline_buf) { @@ -393,11 +518,7 @@ static int fs_path_ensure_buf(struct fs_path *p, int len) if (!tmp_buf) return -ENOMEM; p->buf = tmp_buf; - /* - * The real size of the buffer is bigger, this will let the fast path - * happen most of the time - */ - p->buf_len = ksize(p->buf); + p->buf_len = len; if (p->reversed) { tmp_buf = p->buf + old_buf_len - path_len - 1; @@ -417,12 +538,12 @@ static int fs_path_prepare_for_add(struct fs_path *p, int name_len, int ret; int new_len; - new_len = p->end - p->start + name_len; + new_len = fs_path_len(p) + name_len; if (p->start != p->end) new_len++; ret = fs_path_ensure_buf(p, new_len); if (ret < 0) - goto out; + return ret; if (p->reversed) { if (p->start != p->end) @@ -437,8 +558,7 @@ static int fs_path_prepare_for_add(struct fs_path *p, int name_len, *p->end = 0; } -out: - return ret; + return 0; } static int fs_path_add(struct fs_path *p, const char *name, int name_len) @@ -448,25 +568,15 @@ static int fs_path_add(struct fs_path *p, const char *name, int name_len) ret = fs_path_prepare_for_add(p, name_len, &prepared); if (ret < 0) - goto out; + return ret; memcpy(prepared, name, name_len); -out: - return ret; + return 0; } -static int fs_path_add_path(struct fs_path *p, struct fs_path *p2) +static inline int fs_path_add_path(struct fs_path *p, const struct fs_path *p2) { - int ret; - char *prepared; - - ret = fs_path_prepare_for_add(p, p2->end - p2->start, &prepared); - if (ret < 0) - goto out; - memcpy(prepared, p2->start, p2->end - p2->start); - -out: - return ret; + return fs_path_add(p, p2->start, fs_path_len(p2)); } static int fs_path_add_from_extent_buffer(struct fs_path *p, @@ -478,27 +588,21 @@ static int fs_path_add_from_extent_buffer(struct fs_path *p, ret = fs_path_prepare_for_add(p, len, &prepared); if (ret < 0) - goto out; + return ret; read_extent_buffer(eb, prepared, off, len); -out: - return ret; + return 0; } static int fs_path_copy(struct fs_path *p, struct fs_path *from) { - int ret; - p->reversed = from->reversed; fs_path_reset(p); - ret = fs_path_add_path(p, from); - - return ret; + return fs_path_add_path(p, from); } - static void fs_path_unreverse(struct fs_path *p) { char *tmp; @@ -508,13 +612,21 @@ static void fs_path_unreverse(struct fs_path *p) return; tmp = p->start; - len = p->end - p->start; + len = fs_path_len(p); p->start = p->buf; p->end = p->start + len; memmove(p->start, tmp, len + 1); p->reversed = 0; } +static inline bool is_current_inode_path(const struct send_ctx *sctx, + const struct fs_path *path) +{ + const struct fs_path *cur = &sctx->cur_inode_path; + + return (strncmp(path->start, cur->start, fs_path_len(cur)) == 0); +} + static struct btrfs_path *alloc_path_for_send(void) { struct btrfs_path *path; @@ -522,9 +634,9 @@ static struct btrfs_path *alloc_path_for_send(void) path = btrfs_alloc_path(); if (!path) return NULL; - path->search_commit_root = 1; - path->skip_locking = 1; - path->need_commit_sem = 1; + path->search_commit_root = true; + path->skip_locking = true; + path->need_commit_sem = true; return path; } @@ -535,15 +647,10 @@ static int write_buf(struct file *filp, const void *buf, u32 len, loff_t *off) while (pos < len) { ret = kernel_write(filp, buf + pos, len - pos, off); - /* TODO handle that correctly */ - /*if (ret == -ERESTARTSYS) { - continue; - }*/ if (ret < 0) return ret; - if (ret == 0) { + if (unlikely(ret == 0)) return -EIO; - } pos += ret; } @@ -556,12 +663,15 @@ static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) int total_len = sizeof(*hdr) + len; int left = sctx->send_max_size - sctx->send_size; + if (WARN_ON_ONCE(sctx->put_data)) + return -EINVAL; + if (unlikely(left < total_len)) return -EOVERFLOW; hdr = (struct btrfs_tlv_header *) (sctx->send_buf + sctx->send_size); - hdr->tlv_type = cpu_to_le16(attr); - hdr->tlv_len = cpu_to_le16(len); + put_unaligned_le16(attr, &hdr->tlv_type); + put_unaligned_le16(len, &hdr->tlv_len); memcpy(hdr + 1, data, len); sctx->send_size += total_len; @@ -576,6 +686,8 @@ static int tlv_put(struct send_ctx *sctx, u16 attr, const void *data, int len) return tlv_put(sctx, attr, &__tmp, sizeof(__tmp)); \ } +TLV_PUT_DEFINE_INT(8) +TLV_PUT_DEFINE_INT(32) TLV_PUT_DEFINE_INT(64) static int tlv_put_string(struct send_ctx *sctx, u16 attr, @@ -629,7 +741,7 @@ static int tlv_put_btrfs_timespec(struct send_ctx *sctx, u16 attr, #define TLV_PUT_PATH(sctx, attrtype, p) \ do { \ ret = tlv_put_string(sctx, attrtype, p->start, \ - p->end - p->start); \ + fs_path_len((p))); \ if (ret < 0) \ goto tlv_put_failure; \ } while(0) @@ -650,9 +762,8 @@ static int send_header(struct send_ctx *sctx) { struct btrfs_stream_header hdr; - strcpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); - hdr.version = cpu_to_le32(BTRFS_SEND_STREAM_VERSION); - + strscpy(hdr.magic, BTRFS_SEND_STREAM_MAGIC); + hdr.version = cpu_to_le32(sctx->proto); return write_buf(sctx->send_filp, &hdr, sizeof(hdr), &sctx->send_off); } @@ -667,11 +778,16 @@ static int begin_cmd(struct send_ctx *sctx, int cmd) if (WARN_ON(!sctx->send_buf)) return -EINVAL; - BUG_ON(sctx->send_size); + if (unlikely(sctx->send_size != 0)) { + btrfs_err(sctx->send_root->fs_info, + "send: command header buffer not empty cmd %d offset %llu", + cmd, sctx->send_off); + return -EINVAL; + } sctx->send_size += sizeof(*hdr); hdr = (struct btrfs_cmd_header *)sctx->send_buf; - hdr->cmd = cpu_to_le16(cmd); + put_unaligned_le16(cmd, &hdr->cmd); return 0; } @@ -683,18 +799,17 @@ static int send_cmd(struct send_ctx *sctx) u32 crc; hdr = (struct btrfs_cmd_header *)sctx->send_buf; - hdr->len = cpu_to_le32(sctx->send_size - sizeof(*hdr)); - hdr->crc = 0; + put_unaligned_le32(sctx->send_size - sizeof(*hdr), &hdr->len); + put_unaligned_le32(0, &hdr->crc); crc = crc32c(0, (unsigned char *)sctx->send_buf, sctx->send_size); - hdr->crc = cpu_to_le32(crc); + put_unaligned_le32(crc, &hdr->crc); ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, &sctx->send_off); - sctx->total_send_size += sctx->send_size; - sctx->cmd_send_size[le16_to_cpu(hdr->cmd)] += sctx->send_size; sctx->send_size = 0; + sctx->put_data = false; return ret; } @@ -705,14 +820,11 @@ static int send_cmd(struct send_ctx *sctx) static int send_rename(struct send_ctx *sctx, struct fs_path *from, struct fs_path *to) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret; - btrfs_debug(fs_info, "send_rename %s -> %s", from->start, to->start); - ret = begin_cmd(sctx, BTRFS_SEND_C_RENAME); if (ret < 0) - goto out; + return ret; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, from); TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_TO, to); @@ -720,7 +832,6 @@ static int send_rename(struct send_ctx *sctx, ret = send_cmd(sctx); tlv_put_failure: -out: return ret; } @@ -730,14 +841,11 @@ out: static int send_link(struct send_ctx *sctx, struct fs_path *path, struct fs_path *lnk) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret; - btrfs_debug(fs_info, "send_link %s -> %s", path->start, lnk->start); - ret = begin_cmd(sctx, BTRFS_SEND_C_LINK); if (ret < 0) - goto out; + return ret; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH_LINK, lnk); @@ -745,7 +853,6 @@ static int send_link(struct send_ctx *sctx, ret = send_cmd(sctx); tlv_put_failure: -out: return ret; } @@ -754,21 +861,17 @@ out: */ static int send_unlink(struct send_ctx *sctx, struct fs_path *path) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret; - btrfs_debug(fs_info, "send_unlink %s", path->start); - ret = begin_cmd(sctx, BTRFS_SEND_C_UNLINK); if (ret < 0) - goto out; + return ret; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); ret = send_cmd(sctx); tlv_put_failure: -out: return ret; } @@ -777,35 +880,46 @@ out: */ static int send_rmdir(struct send_ctx *sctx, struct fs_path *path) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret; - btrfs_debug(fs_info, "send_rmdir %s", path->start); - ret = begin_cmd(sctx, BTRFS_SEND_C_RMDIR); if (ret < 0) - goto out; + return ret; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); ret = send_cmd(sctx); tlv_put_failure: -out: return ret; } +struct btrfs_inode_info { + u64 size; + u64 gen; + u64 mode; + u64 uid; + u64 gid; + u64 rdev; + u64 fileattr; + u64 nlink; +}; + /* * Helper function to retrieve some fields from an inode item. */ -static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path, - u64 ino, u64 *size, u64 *gen, u64 *mode, u64 *uid, - u64 *gid, u64 *rdev) +static int get_inode_info(struct btrfs_root *root, u64 ino, + struct btrfs_inode_info *info) { int ret; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_inode_item *ii; struct btrfs_key key; + path = alloc_path_for_send(); + if (!path) + return -ENOMEM; + key.objectid = ino; key.type = BTRFS_INODE_ITEM_KEY; key.offset = 0; @@ -816,44 +930,40 @@ static int __get_inode_info(struct btrfs_root *root, struct btrfs_path *path, return ret; } + if (!info) + return 0; + ii = btrfs_item_ptr(path->nodes[0], path->slots[0], struct btrfs_inode_item); - if (size) - *size = btrfs_inode_size(path->nodes[0], ii); - if (gen) - *gen = btrfs_inode_generation(path->nodes[0], ii); - if (mode) - *mode = btrfs_inode_mode(path->nodes[0], ii); - if (uid) - *uid = btrfs_inode_uid(path->nodes[0], ii); - if (gid) - *gid = btrfs_inode_gid(path->nodes[0], ii); - if (rdev) - *rdev = btrfs_inode_rdev(path->nodes[0], ii); + info->size = btrfs_inode_size(path->nodes[0], ii); + info->gen = btrfs_inode_generation(path->nodes[0], ii); + info->mode = btrfs_inode_mode(path->nodes[0], ii); + info->uid = btrfs_inode_uid(path->nodes[0], ii); + info->gid = btrfs_inode_gid(path->nodes[0], ii); + info->rdev = btrfs_inode_rdev(path->nodes[0], ii); + info->nlink = btrfs_inode_nlink(path->nodes[0], ii); + /* + * Transfer the unchanged u64 value of btrfs_inode_item::flags, that's + * otherwise logically split to 32/32 parts. + */ + info->fileattr = btrfs_inode_flags(path->nodes[0], ii); - return ret; + return 0; } -static int get_inode_info(struct btrfs_root *root, - u64 ino, u64 *size, u64 *gen, - u64 *mode, u64 *uid, u64 *gid, - u64 *rdev) +static int get_inode_gen(struct btrfs_root *root, u64 ino, u64 *gen) { - struct btrfs_path *path; int ret; + struct btrfs_inode_info info = { 0 }; - path = alloc_path_for_send(); - if (!path) - return -ENOMEM; - ret = __get_inode_info(root, path, ino, size, gen, mode, uid, gid, - rdev); - btrfs_free_path(path); + ASSERT(gen); + + ret = get_inode_info(root, ino, &info); + *gen = info.gen; return ret; } -typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, - struct fs_path *p, - void *ctx); +typedef int (*iterate_inode_ref_t)(u64 dir, struct fs_path *p, void *ctx); /* * Helper function to iterate the entries in ONE btrfs_inode_ref or @@ -864,14 +974,13 @@ typedef int (*iterate_inode_ref_t)(int num, u64 dir, int index, * path must point to the INODE_REF or INODE_EXTREF when called. */ static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, - struct btrfs_key *found_key, int resolve, + struct btrfs_key *found_key, bool resolve, iterate_inode_ref_t iterate, void *ctx) { struct extent_buffer *eb = path->nodes[0]; - struct btrfs_item *item; struct btrfs_inode_ref *iref; struct btrfs_inode_extref *extref; - struct btrfs_path *tmp_path; + BTRFS_PATH_AUTO_FREE(tmp_path); struct fs_path *p; u32 cur = 0; u32 total; @@ -879,8 +988,6 @@ static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, u32 name_len; char *start; int ret = 0; - int num = 0; - int index; u64 dir; unsigned long name_off; unsigned long elem_size; @@ -900,12 +1007,11 @@ static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, if (found_key->type == BTRFS_INODE_REF_KEY) { ptr = (unsigned long)btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); - item = btrfs_item_nr(slot); - total = btrfs_item_size(eb, item); + total = btrfs_item_size(eb, slot); elem_size = sizeof(*iref); } else { ptr = btrfs_item_ptr_offset(eb, slot); - total = btrfs_item_size_nr(eb, slot); + total = btrfs_item_size(eb, slot); elem_size = sizeof(*extref); } @@ -916,13 +1022,11 @@ static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, iref = (struct btrfs_inode_ref *)(ptr + cur); name_len = btrfs_inode_ref_name_len(eb, iref); name_off = (unsigned long)(iref + 1); - index = btrfs_inode_ref_index(eb, iref); dir = found_key->offset; } else { extref = (struct btrfs_inode_extref *)(ptr + cur); name_len = btrfs_inode_extref_name_len(eb, extref); name_off = (unsigned long)&extref->name; - index = btrfs_inode_extref_index(eb, extref); dir = btrfs_inode_extref_parent(eb, extref); } @@ -948,7 +1052,13 @@ static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, ret = PTR_ERR(start); goto out; } - BUG_ON(start < p->buf); + if (unlikely(start < p->buf)) { + btrfs_err(root->fs_info, + "send: path ref buffer underflow for key " BTRFS_KEY_FMT, + BTRFS_KEY_FMT_VALUE(found_key)); + ret = -EINVAL; + goto out; + } } p->start = start; } else { @@ -959,14 +1069,12 @@ static int iterate_inode_ref(struct btrfs_root *root, struct btrfs_path *path, } cur += elem_size + name_len; - ret = iterate(num, dir, index, p, ctx); + ret = iterate(dir, p, ctx); if (ret) goto out; - num++; } out: - btrfs_free_path(tmp_path); fs_path_free(p); return ret; } @@ -974,7 +1082,7 @@ out: typedef int (*iterate_dir_item_t)(int num, struct btrfs_key *di_key, const char *name, int name_len, const char *data, int data_len, - u8 type, void *ctx); + void *ctx); /* * Helper function to iterate the entries in ONE btrfs_dir_item. @@ -988,7 +1096,6 @@ static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, { int ret = 0; struct extent_buffer *eb; - struct btrfs_item *item; struct btrfs_dir_item *di; struct btrfs_key di_key; char *buf = NULL; @@ -1000,12 +1107,11 @@ static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, u32 total; int slot; int num; - u8 type; /* * Start with a small buffer (1 page). If later we end up needing more * space, which can happen for xattrs on a fs with a leaf size greater - * then the page size, attempt to increase the buffer. Typically xattr + * than the page size, attempt to increase the buffer. Typically xattr * values are small. */ buf_len = PATH_MAX; @@ -1017,26 +1123,24 @@ static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, eb = path->nodes[0]; slot = path->slots[0]; - item = btrfs_item_nr(slot); di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); cur = 0; len = 0; - total = btrfs_item_size(eb, item); + total = btrfs_item_size(eb, slot); num = 0; while (cur < total) { name_len = btrfs_dir_name_len(eb, di); data_len = btrfs_dir_data_len(eb, di); - type = btrfs_dir_type(eb, di); btrfs_dir_item_key_to_cpu(eb, di, &di_key); - if (type == BTRFS_FT_XATTR) { - if (name_len > XATTR_NAME_MAX) { + if (btrfs_dir_ftype(eb, di) == BTRFS_FT_XATTR) { + if (unlikely(name_len > XATTR_NAME_MAX)) { ret = -ENAMETOOLONG; goto out; } - if (name_len + data_len > - BTRFS_MAX_XATTR_SIZE(root->fs_info)) { + if (unlikely(name_len + data_len > + BTRFS_MAX_XATTR_SIZE(root->fs_info))) { ret = -E2BIG; goto out; } @@ -1044,7 +1148,7 @@ static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, /* * Path too long */ - if (name_len + data_len > PATH_MAX) { + if (unlikely(name_len + data_len > PATH_MAX)) { ret = -ENAMETOOLONG; goto out; } @@ -1080,7 +1184,7 @@ static int iterate_dir_item(struct btrfs_root *root, struct btrfs_path *path, cur += len; ret = iterate(num, &di_key, buf, name_len, buf + name_len, - data_len, type, ctx); + data_len, ctx); if (ret < 0) goto out; if (ret) { @@ -1096,8 +1200,7 @@ out: return ret; } -static int __copy_first_ref(int num, u64 dir, int index, - struct fs_path *p, void *ctx) +static int __copy_first_ref(u64 dir, struct fs_path *p, void *ctx) { int ret; struct fs_path *pt = ctx; @@ -1119,7 +1222,7 @@ static int get_inode_path(struct btrfs_root *root, { int ret; struct btrfs_key key, found_key; - struct btrfs_path *p; + BTRFS_PATH_AUTO_FREE(p); p = alloc_path_for_send(); if (!p) @@ -1133,34 +1236,25 @@ static int get_inode_path(struct btrfs_root *root, ret = btrfs_search_slot_for_read(root, &key, p, 1, 0); if (ret < 0) - goto out; - if (ret) { - ret = 1; - goto out; - } + return ret; + if (ret) + return 1; + btrfs_item_key_to_cpu(p->nodes[0], &found_key, p->slots[0]); if (found_key.objectid != ino || (found_key.type != BTRFS_INODE_REF_KEY && - found_key.type != BTRFS_INODE_EXTREF_KEY)) { - ret = -ENOENT; - goto out; - } + found_key.type != BTRFS_INODE_EXTREF_KEY)) + return -ENOENT; - ret = iterate_inode_ref(root, p, &found_key, 1, - __copy_first_ref, path); + ret = iterate_inode_ref(root, p, &found_key, true, __copy_first_ref, path); if (ret < 0) - goto out; - ret = 0; - -out: - btrfs_free_path(p); - return ret; + return ret; + return 0; } struct backref_ctx { struct send_ctx *sctx; - struct btrfs_path *path; /* number of total found references */ u64 found; @@ -1174,106 +1268,272 @@ struct backref_ctx { /* may be truncated in case it's the last extent in a file */ u64 extent_len; - /* data offset in the file extent item */ - u64 data_offset; - - /* Just to check for bugs in backref resolving */ - int found_itself; + /* The bytenr the file extent item we are processing refers to. */ + u64 bytenr; + /* The owner (root id) of the data backref for the current extent. */ + u64 backref_owner; + /* The offset of the data backref for the current extent. */ + u64 backref_offset; }; static int __clone_root_cmp_bsearch(const void *key, const void *elt) { u64 root = (u64)(uintptr_t)key; - struct clone_root *cr = (struct clone_root *)elt; + const struct clone_root *cr = elt; - if (root < cr->root->root_key.objectid) + if (root < btrfs_root_id(cr->root)) return -1; - if (root > cr->root->root_key.objectid) + if (root > btrfs_root_id(cr->root)) return 1; return 0; } static int __clone_root_cmp_sort(const void *e1, const void *e2) { - struct clone_root *cr1 = (struct clone_root *)e1; - struct clone_root *cr2 = (struct clone_root *)e2; + const struct clone_root *cr1 = e1; + const struct clone_root *cr2 = e2; - if (cr1->root->root_key.objectid < cr2->root->root_key.objectid) + if (btrfs_root_id(cr1->root) < btrfs_root_id(cr2->root)) return -1; - if (cr1->root->root_key.objectid > cr2->root->root_key.objectid) + if (btrfs_root_id(cr1->root) > btrfs_root_id(cr2->root)) return 1; return 0; } /* * Called for every backref that is found for the current extent. - * Results are collected in sctx->clone_roots->ino/offset/found_refs + * Results are collected in sctx->clone_roots->ino/offset. */ -static int __iterate_backrefs(u64 ino, u64 offset, u64 root, void *ctx_) +static int iterate_backrefs(u64 ino, u64 offset, u64 num_bytes, u64 root_id, + void *ctx_) { struct backref_ctx *bctx = ctx_; - struct clone_root *found; - int ret; - u64 i_size; + struct clone_root *clone_root; /* First check if the root is in the list of accepted clone sources */ - found = bsearch((void *)(uintptr_t)root, bctx->sctx->clone_roots, - bctx->sctx->clone_roots_cnt, - sizeof(struct clone_root), - __clone_root_cmp_bsearch); - if (!found) + clone_root = bsearch((void *)(uintptr_t)root_id, bctx->sctx->clone_roots, + bctx->sctx->clone_roots_cnt, + sizeof(struct clone_root), + __clone_root_cmp_bsearch); + if (!clone_root) return 0; - if (found->root == bctx->sctx->send_root && + /* This is our own reference, bail out as we can't clone from it. */ + if (clone_root->root == bctx->sctx->send_root && ino == bctx->cur_objectid && - offset == bctx->cur_offset) { - bctx->found_itself = 1; - } - - /* - * There are inodes that have extents that lie behind its i_size. Don't - * accept clones from these extents. - */ - ret = __get_inode_info(found->root, bctx->path, ino, &i_size, NULL, NULL, - NULL, NULL, NULL); - btrfs_release_path(bctx->path); - if (ret < 0) - return ret; - - if (offset + bctx->data_offset + bctx->extent_len > i_size) + offset == bctx->cur_offset) return 0; /* * Make sure we don't consider clones from send_root that are * behind the current inode/offset. */ - if (found->root == bctx->sctx->send_root) { + if (clone_root->root == bctx->sctx->send_root) { /* - * TODO for the moment we don't accept clones from the inode - * that is currently send. We may change this when - * BTRFS_IOC_CLONE_RANGE supports cloning from and to the same - * file. + * If the source inode was not yet processed we can't issue a + * clone operation, as the source extent does not exist yet at + * the destination of the stream. */ - if (ino >= bctx->cur_objectid) + if (ino > bctx->cur_objectid) + return 0; + /* + * We clone from the inode currently being sent as long as the + * source extent is already processed, otherwise we could try + * to clone from an extent that does not exist yet at the + * destination of the stream. + */ + if (ino == bctx->cur_objectid && + offset + bctx->extent_len > + bctx->sctx->cur_inode_next_write_offset) return 0; } bctx->found++; - found->found_refs++; - if (ino < found->ino) { - found->ino = ino; - found->offset = offset; - } else if (found->ino == ino) { + clone_root->found_ref = true; + + /* + * If the given backref refers to a file extent item with a larger + * number of bytes than what we found before, use the new one so that + * we clone more optimally and end up doing less writes and getting + * less exclusive, non-shared extents at the destination. + */ + if (num_bytes > clone_root->num_bytes) { + clone_root->ino = ino; + clone_root->offset = offset; + clone_root->num_bytes = num_bytes; + + /* + * Found a perfect candidate, so there's no need to continue + * backref walking. + */ + if (num_bytes >= bctx->extent_len) + return BTRFS_ITERATE_EXTENT_INODES_STOP; + } + + return 0; +} + +static bool lookup_backref_cache(u64 leaf_bytenr, void *ctx, + const u64 **root_ids_ret, int *root_count_ret) +{ + struct backref_ctx *bctx = ctx; + struct send_ctx *sctx = bctx->sctx; + struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; + const u64 key = leaf_bytenr >> fs_info->nodesize_bits; + struct btrfs_lru_cache_entry *raw_entry; + struct backref_cache_entry *entry; + + if (sctx->backref_cache.size == 0) + return false; + + /* + * If relocation happened since we first filled the cache, then we must + * empty the cache and can not use it, because even though we operate on + * read-only roots, their leaves and nodes may have been reallocated and + * now be used for different nodes/leaves of the same tree or some other + * tree. + * + * We are called from iterate_extent_inodes() while either holding a + * transaction handle or holding fs_info->commit_root_sem, so no need + * to take any lock here. + */ + if (fs_info->last_reloc_trans > sctx->backref_cache_last_reloc_trans) { + btrfs_lru_cache_clear(&sctx->backref_cache); + return false; + } + + raw_entry = btrfs_lru_cache_lookup(&sctx->backref_cache, key, 0); + if (!raw_entry) + return false; + + entry = container_of(raw_entry, struct backref_cache_entry, entry); + *root_ids_ret = entry->root_ids; + *root_count_ret = entry->num_roots; + + return true; +} + +static void store_backref_cache(u64 leaf_bytenr, const struct ulist *root_ids, + void *ctx) +{ + struct backref_ctx *bctx = ctx; + struct send_ctx *sctx = bctx->sctx; + struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; + struct backref_cache_entry *new_entry; + struct ulist_iterator uiter; + struct ulist_node *node; + int ret; + + /* + * We're called while holding a transaction handle or while holding + * fs_info->commit_root_sem (at iterate_extent_inodes()), so must do a + * NOFS allocation. + */ + new_entry = kmalloc(sizeof(struct backref_cache_entry), GFP_NOFS); + /* No worries, cache is optional. */ + if (!new_entry) + return; + + new_entry->entry.key = leaf_bytenr >> fs_info->nodesize_bits; + new_entry->entry.gen = 0; + new_entry->num_roots = 0; + ULIST_ITER_INIT(&uiter); + while ((node = ulist_next(root_ids, &uiter)) != NULL) { + const u64 root_id = node->val; + struct clone_root *root; + + root = bsearch((void *)(uintptr_t)root_id, sctx->clone_roots, + sctx->clone_roots_cnt, sizeof(struct clone_root), + __clone_root_cmp_bsearch); + if (!root) + continue; + + /* Too many roots, just exit, no worries as caching is optional. */ + if (new_entry->num_roots >= SEND_MAX_BACKREF_CACHE_ROOTS) { + kfree(new_entry); + return; + } + + new_entry->root_ids[new_entry->num_roots] = root_id; + new_entry->num_roots++; + } + + /* + * We may have not added any roots to the new cache entry, which means + * none of the roots is part of the list of roots from which we are + * allowed to clone. Cache the new entry as it's still useful to avoid + * backref walking to determine which roots have a path to the leaf. + * + * Also use GFP_NOFS because we're called while holding a transaction + * handle or while holding fs_info->commit_root_sem. + */ + ret = btrfs_lru_cache_store(&sctx->backref_cache, &new_entry->entry, + GFP_NOFS); + ASSERT(ret == 0 || ret == -ENOMEM); + if (ret) { + /* Caching is optional, no worries. */ + kfree(new_entry); + return; + } + + /* + * We are called from iterate_extent_inodes() while either holding a + * transaction handle or holding fs_info->commit_root_sem, so no need + * to take any lock here. + */ + if (sctx->backref_cache.size == 1) + sctx->backref_cache_last_reloc_trans = fs_info->last_reloc_trans; +} + +static int check_extent_item(u64 bytenr, const struct btrfs_extent_item *ei, + const struct extent_buffer *leaf, void *ctx) +{ + const u64 refs = btrfs_extent_refs(leaf, ei); + const struct backref_ctx *bctx = ctx; + const struct send_ctx *sctx = bctx->sctx; + + if (bytenr == bctx->bytenr) { + const u64 flags = btrfs_extent_flags(leaf, ei); + + if (WARN_ON(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) + return -EUCLEAN; + /* - * same extent found more then once in the same file. + * If we have only one reference and only the send root as a + * clone source - meaning no clone roots were given in the + * struct btrfs_ioctl_send_args passed to the send ioctl - then + * it's our reference and there's no point in doing backref + * walking which is expensive, so exit early. */ - if (found->offset > offset + bctx->extent_len) - found->offset = offset; + if (refs == 1 && sctx->clone_roots_cnt == 1) + return -ENOENT; } + /* + * Backreference walking (iterate_extent_inodes() below) is currently + * too expensive when an extent has a large number of references, both + * in time spent and used memory. So for now just fallback to write + * operations instead of clone operations when an extent has more than + * a certain amount of references. + */ + if (refs > SEND_MAX_EXTENT_REFS) + return -ENOENT; + return 0; } +static bool skip_self_data_ref(u64 root, u64 ino, u64 offset, void *ctx) +{ + const struct backref_ctx *bctx = ctx; + + if (ino == bctx->cur_objectid && + root == bctx->backref_owner && + offset == bctx->backref_offset) + return true; + + return false; +} + /* * Given an inode, offset and extent item, it finds a good clone for a clone * instruction. Returns -ENOENT when none could be found. The function makes @@ -1292,74 +1552,36 @@ static int find_extent_clone(struct send_ctx *sctx, struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret; int extent_type; - u64 logical; u64 disk_byte; u64 num_bytes; - u64 extent_item_pos; - u64 flags = 0; struct btrfs_file_extent_item *fi; struct extent_buffer *eb = path->nodes[0]; - struct backref_ctx *backref_ctx = NULL; + struct backref_ctx backref_ctx = { 0 }; + struct btrfs_backref_walk_ctx backref_walk_ctx = { 0 }; struct clone_root *cur_clone_root; - struct btrfs_key found_key; - struct btrfs_path *tmp_path; int compressed; u32 i; - tmp_path = alloc_path_for_send(); - if (!tmp_path) - return -ENOMEM; - - /* We only use this path under the commit sem */ - tmp_path->need_commit_sem = 0; - - backref_ctx = kmalloc(sizeof(*backref_ctx), GFP_KERNEL); - if (!backref_ctx) { - ret = -ENOMEM; - goto out; - } - - backref_ctx->path = tmp_path; - - if (data_offset >= ino_size) { - /* - * There may be extents that lie behind the file's size. - * I at least had this in combination with snapshotting while - * writing large files. - */ - ret = 0; - goto out; - } + /* + * With fallocate we can get prealloc extents beyond the inode's i_size, + * so we don't do anything here because clone operations can not clone + * to a range beyond i_size without increasing the i_size of the + * destination inode. + */ + if (data_offset >= ino_size) + return 0; - fi = btrfs_item_ptr(eb, path->slots[0], - struct btrfs_file_extent_item); + fi = btrfs_item_ptr(eb, path->slots[0], struct btrfs_file_extent_item); extent_type = btrfs_file_extent_type(eb, fi); - if (extent_type == BTRFS_FILE_EXTENT_INLINE) { - ret = -ENOENT; - goto out; - } - compressed = btrfs_file_extent_compression(eb, fi); + if (extent_type == BTRFS_FILE_EXTENT_INLINE) + return -ENOENT; - num_bytes = btrfs_file_extent_num_bytes(eb, fi); disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); - if (disk_byte == 0) { - ret = -ENOENT; - goto out; - } - logical = disk_byte + btrfs_file_extent_offset(eb, fi); - - down_read(&fs_info->commit_root_sem); - ret = extent_from_logical(fs_info, disk_byte, tmp_path, - &found_key, &flags); - up_read(&fs_info->commit_root_sem); - btrfs_release_path(tmp_path); + if (disk_byte == 0) + return -ENOENT; - if (ret < 0) - goto out; - if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { - ret = -EIO; - goto out; - } + compressed = btrfs_file_extent_compression(eb, fi); + num_bytes = btrfs_file_extent_num_bytes(eb, fi); /* * Setup the clone roots. @@ -1368,77 +1590,105 @@ static int find_extent_clone(struct send_ctx *sctx, cur_clone_root = sctx->clone_roots + i; cur_clone_root->ino = (u64)-1; cur_clone_root->offset = 0; - cur_clone_root->found_refs = 0; + cur_clone_root->num_bytes = 0; + cur_clone_root->found_ref = false; } - backref_ctx->sctx = sctx; - backref_ctx->found = 0; - backref_ctx->cur_objectid = ino; - backref_ctx->cur_offset = data_offset; - backref_ctx->found_itself = 0; - backref_ctx->extent_len = num_bytes; + backref_ctx.sctx = sctx; + backref_ctx.cur_objectid = ino; + backref_ctx.cur_offset = data_offset; + backref_ctx.bytenr = disk_byte; /* - * For non-compressed extents iterate_extent_inodes() gives us extent - * offsets that already take into account the data offset, but not for - * compressed extents, since the offset is logical and not relative to - * the physical extent locations. We must take this into account to - * avoid sending clone offsets that go beyond the source file's size, - * which would result in the clone ioctl failing with -EINVAL on the - * receiving end. + * Use the header owner and not the send root's id, because in case of a + * snapshot we can have shared subtrees. */ - if (compressed == BTRFS_COMPRESS_NONE) - backref_ctx->data_offset = 0; - else - backref_ctx->data_offset = btrfs_file_extent_offset(eb, fi); + backref_ctx.backref_owner = btrfs_header_owner(eb); + backref_ctx.backref_offset = data_offset - btrfs_file_extent_offset(eb, fi); /* * The last extent of a file may be too large due to page alignment. * We need to adjust extent_len in this case so that the checks in - * __iterate_backrefs work. + * iterate_backrefs() work. */ if (data_offset + num_bytes >= ino_size) - backref_ctx->extent_len = ino_size - data_offset; + backref_ctx.extent_len = ino_size - data_offset; + else + backref_ctx.extent_len = num_bytes; /* * Now collect all backrefs. */ + backref_walk_ctx.bytenr = disk_byte; if (compressed == BTRFS_COMPRESS_NONE) - extent_item_pos = logical - found_key.objectid; - else - extent_item_pos = 0; - ret = iterate_extent_inodes(fs_info, found_key.objectid, - extent_item_pos, 1, __iterate_backrefs, - backref_ctx, false); + backref_walk_ctx.extent_item_pos = btrfs_file_extent_offset(eb, fi); + backref_walk_ctx.fs_info = fs_info; + backref_walk_ctx.cache_lookup = lookup_backref_cache; + backref_walk_ctx.cache_store = store_backref_cache; + backref_walk_ctx.indirect_ref_iterator = iterate_backrefs; + backref_walk_ctx.check_extent_item = check_extent_item; + backref_walk_ctx.user_ctx = &backref_ctx; + /* + * If have a single clone root, then it's the send root and we can tell + * the backref walking code to skip our own backref and not resolve it, + * since we can not use it for cloning - the source and destination + * ranges can't overlap and in case the leaf is shared through a subtree + * due to snapshots, we can't use those other roots since they are not + * in the list of clone roots. + */ + if (sctx->clone_roots_cnt == 1) + backref_walk_ctx.skip_data_ref = skip_self_data_ref; + + ret = iterate_extent_inodes(&backref_walk_ctx, true, iterate_backrefs, + &backref_ctx); if (ret < 0) - goto out; + return ret; - if (!backref_ctx->found_itself) { - /* found a bug in backref code? */ - ret = -EIO; - btrfs_err(fs_info, - "did not find backref in send_root. inode=%llu, offset=%llu, disk_byte=%llu found extent=%llu", - ino, data_offset, disk_byte, found_key.objectid); - goto out; + down_read(&fs_info->commit_root_sem); + if (fs_info->last_reloc_trans > sctx->last_reloc_trans) { + /* + * A transaction commit for a transaction in which block group + * relocation was done just happened. + * The disk_bytenr of the file extent item we processed is + * possibly stale, referring to the extent's location before + * relocation. So act as if we haven't found any clone sources + * and fallback to write commands, which will read the correct + * data from the new extent location. Otherwise we will fail + * below because we haven't found our own back reference or we + * could be getting incorrect sources in case the old extent + * was already reallocated after the relocation. + */ + up_read(&fs_info->commit_root_sem); + return -ENOENT; } + up_read(&fs_info->commit_root_sem); - btrfs_debug(fs_info, - "find_extent_clone: data_offset=%llu, ino=%llu, num_bytes=%llu, logical=%llu", - data_offset, ino, num_bytes, logical); - - if (!backref_ctx->found) - btrfs_debug(fs_info, "no clones found"); + if (!backref_ctx.found) + return -ENOENT; cur_clone_root = NULL; for (i = 0; i < sctx->clone_roots_cnt; i++) { - if (sctx->clone_roots[i].found_refs) { - if (!cur_clone_root) - cur_clone_root = sctx->clone_roots + i; - else if (sctx->clone_roots[i].root == sctx->send_root) - /* prefer clones from send_root over others */ - cur_clone_root = sctx->clone_roots + i; - } + struct clone_root *clone_root = &sctx->clone_roots[i]; + + if (!clone_root->found_ref) + continue; + /* + * Choose the root from which we can clone more bytes, to + * minimize write operations and therefore have more extent + * sharing at the destination (the same as in the source). + */ + if (!cur_clone_root || + clone_root->num_bytes > cur_clone_root->num_bytes) { + cur_clone_root = clone_root; + + /* + * We found an optimal clone candidate (any inode from + * any root is fine), so we're done. + */ + if (clone_root->num_bytes >= backref_ctx.extent_len) + break; + } } if (cur_clone_root) { @@ -1448,9 +1698,6 @@ static int find_extent_clone(struct send_ctx *sctx, ret = -ENOENT; } -out: - btrfs_free_path(tmp_path); - kfree(backref_ctx); return ret; } @@ -1459,7 +1706,7 @@ static int read_symlink(struct btrfs_root *root, struct fs_path *dest) { int ret; - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; struct btrfs_file_extent_item *ei; u8 type; @@ -1476,38 +1723,45 @@ static int read_symlink(struct btrfs_root *root, key.offset = 0; ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); if (ret < 0) - goto out; - if (ret) { + return ret; + if (unlikely(ret)) { /* * An empty symlink inode. Can happen in rare error paths when * creating a symlink (transaction committed before the inode * eviction handler removed the symlink inode items and a crash - * happened in between or the subvol was snapshoted in between). + * happened in between or the subvol was snapshotted in between). * Print an informative message to dmesg/syslog so that the user * can delete the symlink. */ btrfs_err(root->fs_info, "Found empty symlink inode %llu at root %llu", - ino, root->root_key.objectid); - ret = -EIO; - goto out; + ino, btrfs_root_id(root)); + return -EIO; } ei = btrfs_item_ptr(path->nodes[0], path->slots[0], struct btrfs_file_extent_item); type = btrfs_file_extent_type(path->nodes[0], ei); + if (unlikely(type != BTRFS_FILE_EXTENT_INLINE)) { + ret = -EUCLEAN; + btrfs_crit(root->fs_info, +"send: found symlink extent that is not inline, ino %llu root %llu extent type %d", + ino, btrfs_root_id(root), type); + return ret; + } compression = btrfs_file_extent_compression(path->nodes[0], ei); - BUG_ON(type != BTRFS_FILE_EXTENT_INLINE); - BUG_ON(compression); + if (unlikely(compression != BTRFS_COMPRESS_NONE)) { + ret = -EUCLEAN; + btrfs_crit(root->fs_info, +"send: found symlink extent with compression, ino %llu root %llu compression type %d", + ino, btrfs_root_id(root), compression); + return ret; + } off = btrfs_file_extent_inline_start(ei); len = btrfs_file_extent_ram_bytes(path->nodes[0], ei); - ret = fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); - -out: - btrfs_free_path(path); - return ret; + return fs_path_add_from_extent_buffer(dest, path->nodes[0], off, len); } /* @@ -1518,8 +1772,7 @@ static int gen_unique_name(struct send_ctx *sctx, u64 ino, u64 gen, struct fs_path *dest) { - int ret = 0; - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_dir_item *di; char tmp[64]; int len; @@ -1530,18 +1783,21 @@ static int gen_unique_name(struct send_ctx *sctx, return -ENOMEM; while (1) { + struct fscrypt_str tmp_name; + len = snprintf(tmp, sizeof(tmp), "o%llu-%llu-%llu", ino, gen, idx); ASSERT(len < sizeof(tmp)); + tmp_name.name = tmp; + tmp_name.len = len; di = btrfs_lookup_dir_item(NULL, sctx->send_root, path, BTRFS_FIRST_FREE_OBJECTID, - tmp, strlen(tmp), 0); + &tmp_name, 0); btrfs_release_path(path); - if (IS_ERR(di)) { - ret = PTR_ERR(di); - goto out; - } + if (IS_ERR(di)) + return PTR_ERR(di); + if (di) { /* not unique, try again */ idx++; @@ -1550,18 +1806,16 @@ static int gen_unique_name(struct send_ctx *sctx, if (!sctx->parent_root) { /* unique */ - ret = 0; break; } di = btrfs_lookup_dir_item(NULL, sctx->parent_root, path, BTRFS_FIRST_FREE_OBJECTID, - tmp, strlen(tmp), 0); + &tmp_name, 0); btrfs_release_path(path); - if (IS_ERR(di)) { - ret = PTR_ERR(di); - goto out; - } + if (IS_ERR(di)) + return PTR_ERR(di); + if (di) { /* not unique, try again */ idx++; @@ -1571,11 +1825,7 @@ static int gen_unique_name(struct send_ctx *sctx, break; } - ret = fs_path_add(dest, tmp, strlen(tmp)); - -out: - btrfs_free_path(path); - return ret; + return fs_path_add(dest, tmp, len); } enum inode_state { @@ -1586,28 +1836,34 @@ enum inode_state { inode_state_did_delete, }; -static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen) +static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen, + u64 *send_gen, u64 *parent_gen) { int ret; int left_ret; int right_ret; u64 left_gen; - u64 right_gen; + u64 right_gen = 0; + struct btrfs_inode_info info; - ret = get_inode_info(sctx->send_root, ino, NULL, &left_gen, NULL, NULL, - NULL, NULL); + ret = get_inode_info(sctx->send_root, ino, &info); if (ret < 0 && ret != -ENOENT) - goto out; - left_ret = ret; + return ret; + left_ret = (info.nlink == 0) ? -ENOENT : ret; + left_gen = info.gen; + if (send_gen) + *send_gen = ((left_ret == -ENOENT) ? 0 : info.gen); if (!sctx->parent_root) { right_ret = -ENOENT; } else { - ret = get_inode_info(sctx->parent_root, ino, NULL, &right_gen, - NULL, NULL, NULL, NULL); + ret = get_inode_info(sctx->parent_root, ino, &info); if (ret < 0 && ret != -ENOENT) - goto out; - right_ret = ret; + return ret; + right_ret = (info.nlink == 0) ? -ENOENT : ret; + right_gen = info.gen; + if (parent_gen) + *parent_gen = ((right_ret == -ENOENT) ? 0 : info.gen); } if (!left_ret && !right_ret) { @@ -1648,30 +1904,27 @@ static int get_cur_inode_state(struct send_ctx *sctx, u64 ino, u64 gen) ret = -ENOENT; } -out: return ret; } -static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen) +static int is_inode_existent(struct send_ctx *sctx, u64 ino, u64 gen, + u64 *send_gen, u64 *parent_gen) { int ret; if (ino == BTRFS_FIRST_FREE_OBJECTID) return 1; - ret = get_cur_inode_state(sctx, ino, gen); + ret = get_cur_inode_state(sctx, ino, gen, send_gen, parent_gen); if (ret < 0) - goto out; + return ret; if (ret == inode_state_no_change || ret == inode_state_did_create || ret == inode_state_will_delete) - ret = 1; - else - ret = 0; + return 1; -out: - return ret; + return 0; } /* @@ -1679,34 +1932,28 @@ out: */ static int lookup_dir_item_inode(struct btrfs_root *root, u64 dir, const char *name, int name_len, - u64 *found_inode, - u8 *found_type) + u64 *found_inode) { int ret = 0; struct btrfs_dir_item *di; struct btrfs_key key; - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); + struct fscrypt_str name_str = FSTR_INIT((char *)name, name_len); path = alloc_path_for_send(); if (!path) return -ENOMEM; - di = btrfs_lookup_dir_item(NULL, root, path, - dir, name, name_len, 0); - if (IS_ERR_OR_NULL(di)) { - ret = di ? PTR_ERR(di) : -ENOENT; - goto out; - } + di = btrfs_lookup_dir_item(NULL, root, path, dir, &name_str, 0); + if (IS_ERR_OR_NULL(di)) + return di ? PTR_ERR(di) : -ENOENT; + btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); - if (key.type == BTRFS_ROOT_ITEM_KEY) { - ret = -ENOENT; - goto out; - } + if (key.type == BTRFS_ROOT_ITEM_KEY) + return -ENOENT; + *found_inode = key.objectid; - *found_type = btrfs_dir_type(path->nodes[0], di); -out: - btrfs_free_path(path); return ret; } @@ -1720,7 +1967,7 @@ static int get_first_ref(struct btrfs_root *root, u64 ino, int ret; struct btrfs_key key; struct btrfs_key found_key; - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); int len; u64 parent_dir; @@ -1734,16 +1981,14 @@ static int get_first_ref(struct btrfs_root *root, u64 ino, ret = btrfs_search_slot_for_read(root, &key, path, 1, 0); if (ret < 0) - goto out; + return ret; if (!ret) btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); if (ret || found_key.objectid != ino || (found_key.type != BTRFS_INODE_REF_KEY && - found_key.type != BTRFS_INODE_EXTREF_KEY)) { - ret = -ENOENT; - goto out; - } + found_key.type != BTRFS_INODE_EXTREF_KEY)) + return -ENOENT; if (found_key.type == BTRFS_INODE_REF_KEY) { struct btrfs_inode_ref *iref; @@ -1764,20 +2009,17 @@ static int get_first_ref(struct btrfs_root *root, u64 ino, parent_dir = btrfs_inode_extref_parent(path->nodes[0], extref); } if (ret < 0) - goto out; + return ret; btrfs_release_path(path); if (dir_gen) { - ret = get_inode_info(root, parent_dir, NULL, dir_gen, NULL, - NULL, NULL, NULL); + ret = get_inode_gen(root, parent_dir, dir_gen); if (ret < 0) - goto out; + return ret; } *dir = parent_dir; -out: - btrfs_free_path(path); return ret; } @@ -1823,44 +2065,36 @@ static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, const char *name, int name_len, u64 *who_ino, u64 *who_gen, u64 *who_mode) { - int ret = 0; - u64 gen; + int ret; + u64 parent_root_dir_gen; u64 other_inode = 0; - u8 other_type = 0; + struct btrfs_inode_info info; if (!sctx->parent_root) - goto out; + return 0; - ret = is_inode_existent(sctx, dir, dir_gen); + ret = is_inode_existent(sctx, dir, dir_gen, NULL, &parent_root_dir_gen); if (ret <= 0) - goto out; + return 0; /* * If we have a parent root we need to verify that the parent dir was * not deleted and then re-created, if it was then we have no overwrite * and we can just unlink this entry. + * + * @parent_root_dir_gen was set to 0 if the inode does not exist in the + * parent root. */ - if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID) { - ret = get_inode_info(sctx->parent_root, dir, NULL, &gen, NULL, - NULL, NULL, NULL); - if (ret < 0 && ret != -ENOENT) - goto out; - if (ret) { - ret = 0; - goto out; - } - if (gen != dir_gen) - goto out; - } + if (sctx->parent_root && dir != BTRFS_FIRST_FREE_OBJECTID && + parent_root_dir_gen != dir_gen) + return 0; ret = lookup_dir_item_inode(sctx->parent_root, dir, name, name_len, - &other_inode, &other_type); - if (ret < 0 && ret != -ENOENT) - goto out; - if (ret) { - ret = 0; - goto out; - } + &other_inode); + if (ret == -ENOENT) + return 0; + else if (ret < 0) + return ret; /* * Check if the overwritten ref was already processed. If yes, the ref @@ -1869,19 +2103,17 @@ static int will_overwrite_ref(struct send_ctx *sctx, u64 dir, u64 dir_gen, */ if (other_inode > sctx->send_progress || is_waiting_for_move(sctx, other_inode)) { - ret = get_inode_info(sctx->parent_root, other_inode, NULL, - who_gen, who_mode, NULL, NULL, NULL); + ret = get_inode_info(sctx->parent_root, other_inode, &info); if (ret < 0) - goto out; + return ret; - ret = 1; *who_ino = other_inode; - } else { - ret = 0; + *who_gen = info.gen; + *who_mode = info.mode; + return 1; } -out: - return ret; + return 0; } /* @@ -1896,50 +2128,43 @@ static int did_overwrite_ref(struct send_ctx *sctx, u64 ino, u64 ino_gen, const char *name, int name_len) { - int ret = 0; - u64 gen; + int ret; u64 ow_inode; - u8 other_type; + u64 ow_gen = 0; + u64 send_root_dir_gen; if (!sctx->parent_root) - goto out; + return 0; - ret = is_inode_existent(sctx, dir, dir_gen); + ret = is_inode_existent(sctx, dir, dir_gen, &send_root_dir_gen, NULL); if (ret <= 0) - goto out; + return ret; - if (dir != BTRFS_FIRST_FREE_OBJECTID) { - ret = get_inode_info(sctx->send_root, dir, NULL, &gen, NULL, - NULL, NULL, NULL); - if (ret < 0 && ret != -ENOENT) - goto out; - if (ret) { - ret = 0; - goto out; - } - if (gen != dir_gen) - goto out; - } + /* + * @send_root_dir_gen was set to 0 if the inode does not exist in the + * send root. + */ + if (dir != BTRFS_FIRST_FREE_OBJECTID && send_root_dir_gen != dir_gen) + return 0; /* check if the ref was overwritten by another ref */ ret = lookup_dir_item_inode(sctx->send_root, dir, name, name_len, - &ow_inode, &other_type); - if (ret < 0 && ret != -ENOENT) - goto out; - if (ret) { + &ow_inode); + if (ret == -ENOENT) { /* was never and will never be overwritten */ - ret = 0; - goto out; + return 0; + } else if (ret < 0) { + return ret; } - ret = get_inode_info(sctx->send_root, ow_inode, NULL, &gen, NULL, NULL, - NULL, NULL); - if (ret < 0) - goto out; + if (ow_inode == ino) { + ret = get_inode_gen(sctx->send_root, ow_inode, &ow_gen); + if (ret < 0) + return ret; - if (ow_inode == ino && gen == ino_gen) { - ret = 0; - goto out; + /* It's the same inode, so no overwrite happened. */ + if (ow_gen == ino_gen) + return 0; } /* @@ -1948,15 +2173,20 @@ static int did_overwrite_ref(struct send_ctx *sctx, * inode 'ino' to be orphanized, therefore check if ow_inode matches * the current inode being processed. */ - if ((ow_inode < sctx->send_progress) || - (ino != sctx->cur_ino && ow_inode == sctx->cur_ino && - gen == sctx->cur_inode_gen)) - ret = 1; - else - ret = 0; + if (ow_inode < sctx->send_progress) + return 1; -out: - return ret; + if (ino != sctx->cur_ino && ow_inode == sctx->cur_ino) { + if (ow_gen == 0) { + ret = get_inode_gen(sctx->send_root, ow_inode, &ow_gen); + if (ret < 0) + return ret; + } + if (ow_gen == sctx->cur_inode_gen) + return 1; + } + + return 0; } /* @@ -1990,123 +2220,16 @@ out: return ret; } -/* - * Insert a name cache entry. On 32bit kernels the radix tree index is 32bit, - * so we need to do some special handling in case we have clashes. This function - * takes care of this with the help of name_cache_entry::radix_list. - * In case of error, nce is kfreed. - */ -static int name_cache_insert(struct send_ctx *sctx, - struct name_cache_entry *nce) -{ - int ret = 0; - struct list_head *nce_head; - - nce_head = radix_tree_lookup(&sctx->name_cache, - (unsigned long)nce->ino); - if (!nce_head) { - nce_head = kmalloc(sizeof(*nce_head), GFP_KERNEL); - if (!nce_head) { - kfree(nce); - return -ENOMEM; - } - INIT_LIST_HEAD(nce_head); - - ret = radix_tree_insert(&sctx->name_cache, nce->ino, nce_head); - if (ret < 0) { - kfree(nce_head); - kfree(nce); - return ret; - } - } - list_add_tail(&nce->radix_list, nce_head); - list_add_tail(&nce->list, &sctx->name_cache_list); - sctx->name_cache_size++; - - return ret; -} - -static void name_cache_delete(struct send_ctx *sctx, - struct name_cache_entry *nce) +static inline struct name_cache_entry *name_cache_search(struct send_ctx *sctx, + u64 ino, u64 gen) { - struct list_head *nce_head; - - nce_head = radix_tree_lookup(&sctx->name_cache, - (unsigned long)nce->ino); - if (!nce_head) { - btrfs_err(sctx->send_root->fs_info, - "name_cache_delete lookup failed ino %llu cache size %d, leaking memory", - nce->ino, sctx->name_cache_size); - } - - list_del(&nce->radix_list); - list_del(&nce->list); - sctx->name_cache_size--; + struct btrfs_lru_cache_entry *entry; - /* - * We may not get to the final release of nce_head if the lookup fails - */ - if (nce_head && list_empty(nce_head)) { - radix_tree_delete(&sctx->name_cache, (unsigned long)nce->ino); - kfree(nce_head); - } -} - -static struct name_cache_entry *name_cache_search(struct send_ctx *sctx, - u64 ino, u64 gen) -{ - struct list_head *nce_head; - struct name_cache_entry *cur; - - nce_head = radix_tree_lookup(&sctx->name_cache, (unsigned long)ino); - if (!nce_head) + entry = btrfs_lru_cache_lookup(&sctx->name_cache, ino, gen); + if (!entry) return NULL; - list_for_each_entry(cur, nce_head, radix_list) { - if (cur->ino == ino && cur->gen == gen) - return cur; - } - return NULL; -} - -/* - * Removes the entry from the list and adds it back to the end. This marks the - * entry as recently used so that name_cache_clean_unused does not remove it. - */ -static void name_cache_used(struct send_ctx *sctx, struct name_cache_entry *nce) -{ - list_del(&nce->list); - list_add_tail(&nce->list, &sctx->name_cache_list); -} - -/* - * Remove some entries from the beginning of name_cache_list. - */ -static void name_cache_clean_unused(struct send_ctx *sctx) -{ - struct name_cache_entry *nce; - - if (sctx->name_cache_size < SEND_CTX_NAME_CACHE_CLEAN_SIZE) - return; - - while (sctx->name_cache_size > SEND_CTX_MAX_NAME_CACHE_SIZE) { - nce = list_entry(sctx->name_cache_list.next, - struct name_cache_entry, list); - name_cache_delete(sctx, nce); - kfree(nce); - } -} - -static void name_cache_free(struct send_ctx *sctx) -{ - struct name_cache_entry *nce; - - while (!list_empty(&sctx->name_cache_list)) { - nce = list_entry(sctx->name_cache_list.next, - struct name_cache_entry, list); - name_cache_delete(sctx, nce); - kfree(nce); - } + return container_of(entry, struct name_cache_entry, entry); } /* @@ -2125,7 +2248,7 @@ static int __get_cur_name_and_parent(struct send_ctx *sctx, { int ret; int nce_ret; - struct name_cache_entry *nce = NULL; + struct name_cache_entry *nce; /* * First check if we already did a call to this function with the same @@ -2135,34 +2258,31 @@ static int __get_cur_name_and_parent(struct send_ctx *sctx, nce = name_cache_search(sctx, ino, gen); if (nce) { if (ino < sctx->send_progress && nce->need_later_update) { - name_cache_delete(sctx, nce); - kfree(nce); + btrfs_lru_cache_remove(&sctx->name_cache, &nce->entry); nce = NULL; } else { - name_cache_used(sctx, nce); *parent_ino = nce->parent_ino; *parent_gen = nce->parent_gen; ret = fs_path_add(dest, nce->name, nce->name_len); if (ret < 0) - goto out; - ret = nce->ret; - goto out; + return ret; + return nce->ret; } } /* * If the inode is not existent yet, add the orphan name and return 1. * This should only happen for the parent dir that we determine in - * __record_new_ref + * record_new_ref_if_needed(). */ - ret = is_inode_existent(sctx, ino, gen); + ret = is_inode_existent(sctx, ino, gen, NULL, NULL); if (ret < 0) - goto out; + return ret; if (!ret) { ret = gen_unique_name(sctx, ino, gen, dest); if (ret < 0) - goto out; + return ret; ret = 1; goto out_cache; } @@ -2178,21 +2298,21 @@ static int __get_cur_name_and_parent(struct send_ctx *sctx, ret = get_first_ref(sctx->parent_root, ino, parent_ino, parent_gen, dest); if (ret < 0) - goto out; + return ret; /* * Check if the ref was overwritten by an inode's ref that was processed * earlier. If yes, treat as orphan and return 1. */ ret = did_overwrite_ref(sctx, *parent_ino, *parent_gen, ino, gen, - dest->start, dest->end - dest->start); + dest->start, fs_path_len(dest)); if (ret < 0) - goto out; + return ret; if (ret) { fs_path_reset(dest); ret = gen_unique_name(sctx, ino, gen, dest); if (ret < 0) - goto out; + return ret; ret = 1; } @@ -2200,31 +2320,29 @@ out_cache: /* * Store the result of the lookup in the name cache. */ - nce = kmalloc(sizeof(*nce) + fs_path_len(dest) + 1, GFP_KERNEL); - if (!nce) { - ret = -ENOMEM; - goto out; - } + nce = kmalloc(sizeof(*nce) + fs_path_len(dest), GFP_KERNEL); + if (!nce) + return -ENOMEM; - nce->ino = ino; - nce->gen = gen; + nce->entry.key = ino; + nce->entry.gen = gen; nce->parent_ino = *parent_ino; nce->parent_gen = *parent_gen; nce->name_len = fs_path_len(dest); nce->ret = ret; - strcpy(nce->name, dest->start); + memcpy(nce->name, dest->start, nce->name_len); if (ino < sctx->send_progress) nce->need_later_update = 0; else nce->need_later_update = 1; - nce_ret = name_cache_insert(sctx, nce); - if (nce_ret < 0) - ret = nce_ret; - name_cache_clean_unused(sctx); + nce_ret = btrfs_lru_cache_store(&sctx->name_cache, &nce->entry, GFP_KERNEL); + if (nce_ret < 0) { + kfree(nce); + return nce_ret; + } -out: return ret; } @@ -2261,6 +2379,14 @@ static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, u64 parent_inode = 0; u64 parent_gen = 0; int stop = 0; + const bool is_cur_inode = (ino == sctx->cur_ino && gen == sctx->cur_inode_gen); + + if (is_cur_inode && fs_path_len(&sctx->cur_inode_path) > 0) { + if (dest != &sctx->cur_inode_path) + return fs_path_copy(dest, &sctx->cur_inode_path); + + return 0; + } name = fs_path_alloc(); if (!name) { @@ -2276,7 +2402,7 @@ static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, fs_path_reset(name); - if (is_waiting_for_rm(sctx, ino)) { + if (is_waiting_for_rm(sctx, ino, gen)) { ret = gen_unique_name(sctx, ino, gen, name); if (ret < 0) goto out; @@ -2312,8 +2438,12 @@ static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen, out: fs_path_free(name); - if (!ret) + if (!ret) { fs_path_unreverse(dest); + if (is_cur_inode && dest != &sctx->cur_inode_path) + ret = fs_path_copy(&sctx->cur_inode_path, dest); + } + return ret; } @@ -2325,11 +2455,11 @@ static int send_subvol_begin(struct send_ctx *sctx) int ret; struct btrfs_root *send_root = sctx->send_root; struct btrfs_root *parent_root = sctx->parent_root; - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; struct btrfs_root_ref *ref; struct extent_buffer *leaf; - char *name = NULL; + char AUTO_KFREE(name); int namelen; path = btrfs_alloc_path(); @@ -2337,30 +2467,25 @@ static int send_subvol_begin(struct send_ctx *sctx) return -ENOMEM; name = kmalloc(BTRFS_PATH_NAME_MAX, GFP_KERNEL); - if (!name) { - btrfs_free_path(path); + if (!name) return -ENOMEM; - } - key.objectid = send_root->root_key.objectid; + key.objectid = btrfs_root_id(send_root); key.type = BTRFS_ROOT_BACKREF_KEY; key.offset = 0; ret = btrfs_search_slot_for_read(send_root->fs_info->tree_root, &key, path, 1, 0); if (ret < 0) - goto out; - if (ret) { - ret = -ENOENT; - goto out; - } + return ret; + if (ret) + return -ENOENT; leaf = path->nodes[0]; btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); if (key.type != BTRFS_ROOT_BACKREF_KEY || - key.objectid != send_root->root_key.objectid) { - ret = -ENOENT; - goto out; + key.objectid != btrfs_root_id(send_root)) { + return -ENOENT; } ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref); namelen = btrfs_root_ref_name_len(leaf, ref); @@ -2370,11 +2495,11 @@ static int send_subvol_begin(struct send_ctx *sctx) if (parent_root) { ret = begin_cmd(sctx, BTRFS_SEND_C_SNAPSHOT); if (ret < 0) - goto out; + return ret; } else { ret = begin_cmd(sctx, BTRFS_SEND_C_SUBVOL); if (ret < 0) - goto out; + return ret; } TLV_PUT_STRING(sctx, BTRFS_SEND_A_PATH, name, namelen); @@ -2387,7 +2512,7 @@ static int send_subvol_begin(struct send_ctx *sctx) sctx->send_root->root_item.uuid); TLV_PUT_U64(sctx, BTRFS_SEND_A_CTRANSID, - le64_to_cpu(sctx->send_root->root_item.ctransid)); + btrfs_root_ctransid(&sctx->send_root->root_item)); if (parent_root) { if (!btrfs_is_empty_uuid(parent_root->root_item.received_uuid)) TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, @@ -2396,37 +2521,69 @@ static int send_subvol_begin(struct send_ctx *sctx) TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, parent_root->root_item.uuid); TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, - le64_to_cpu(sctx->parent_root->root_item.ctransid)); + btrfs_root_ctransid(&sctx->parent_root->root_item)); } ret = send_cmd(sctx); tlv_put_failure: -out: - btrfs_free_path(path); - kfree(name); return ret; } +static struct fs_path *get_cur_inode_path(struct send_ctx *sctx) +{ + if (fs_path_len(&sctx->cur_inode_path) == 0) { + int ret; + + ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, + &sctx->cur_inode_path); + if (ret < 0) + return ERR_PTR(ret); + } + + return &sctx->cur_inode_path; +} + +static struct fs_path *get_path_for_command(struct send_ctx *sctx, u64 ino, u64 gen) +{ + struct fs_path *path; + int ret; + + if (ino == sctx->cur_ino && gen == sctx->cur_inode_gen) + return get_cur_inode_path(sctx); + + path = fs_path_alloc(); + if (!path) + return ERR_PTR(-ENOMEM); + + ret = get_cur_path(sctx, ino, gen, path); + if (ret < 0) { + fs_path_free(path); + return ERR_PTR(ret); + } + + return path; +} + +static void free_path_for_command(const struct send_ctx *sctx, struct fs_path *path) +{ + if (path != &sctx->cur_inode_path) + fs_path_free(path); +} + static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret = 0; struct fs_path *p; - btrfs_debug(fs_info, "send_truncate %llu size=%llu", ino, size); - - p = fs_path_alloc(); - if (!p) - return -ENOMEM; + p = get_path_for_command(sctx, ino, gen); + if (IS_ERR(p)) + return PTR_ERR(p); ret = begin_cmd(sctx, BTRFS_SEND_C_TRUNCATE); if (ret < 0) goto out; - ret = get_cur_path(sctx, ino, gen, p); - if (ret < 0) - goto out; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, size); @@ -2434,29 +2591,23 @@ static int send_truncate(struct send_ctx *sctx, u64 ino, u64 gen, u64 size) tlv_put_failure: out: - fs_path_free(p); + free_path_for_command(sctx, p); return ret; } static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret = 0; struct fs_path *p; - btrfs_debug(fs_info, "send_chmod %llu mode=%llu", ino, mode); - - p = fs_path_alloc(); - if (!p) - return -ENOMEM; + p = get_path_for_command(sctx, ino, gen); + if (IS_ERR(p)) + return PTR_ERR(p); ret = begin_cmd(sctx, BTRFS_SEND_C_CHMOD); if (ret < 0) goto out; - ret = get_cur_path(sctx, ino, gen, p); - if (ret < 0) - goto out; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); TLV_PUT_U64(sctx, BTRFS_SEND_A_MODE, mode & 07777); @@ -2464,30 +2615,50 @@ static int send_chmod(struct send_ctx *sctx, u64 ino, u64 gen, u64 mode) tlv_put_failure: out: - fs_path_free(p); + free_path_for_command(sctx, p); return ret; } -static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) +static int send_fileattr(struct send_ctx *sctx, u64 ino, u64 gen, u64 fileattr) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret = 0; struct fs_path *p; - btrfs_debug(fs_info, "send_chown %llu uid=%llu, gid=%llu", - ino, uid, gid); + if (sctx->proto < 2) + return 0; - p = fs_path_alloc(); - if (!p) - return -ENOMEM; + p = get_path_for_command(sctx, ino, gen); + if (IS_ERR(p)) + return PTR_ERR(p); - ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); + ret = begin_cmd(sctx, BTRFS_SEND_C_FILEATTR); if (ret < 0) goto out; - ret = get_cur_path(sctx, ino, gen, p); + TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); + TLV_PUT_U64(sctx, BTRFS_SEND_A_FILEATTR, fileattr); + + ret = send_cmd(sctx); + +tlv_put_failure: +out: + free_path_for_command(sctx, p); + return ret; +} + +static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) +{ + int ret = 0; + struct fs_path *p; + + p = get_path_for_command(sctx, ino, gen); + if (IS_ERR(p)) + return PTR_ERR(p); + + ret = begin_cmd(sctx, BTRFS_SEND_C_CHOWN); if (ret < 0) goto out; + TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); TLV_PUT_U64(sctx, BTRFS_SEND_A_UID, uid); TLV_PUT_U64(sctx, BTRFS_SEND_A_GID, gid); @@ -2496,26 +2667,23 @@ static int send_chown(struct send_ctx *sctx, u64 ino, u64 gen, u64 uid, u64 gid) tlv_put_failure: out: - fs_path_free(p); + free_path_for_command(sctx, p); return ret; } static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret = 0; struct fs_path *p = NULL; struct btrfs_inode_item *ii; - struct btrfs_path *path = NULL; + BTRFS_PATH_AUTO_FREE(path); struct extent_buffer *eb; struct btrfs_key key; int slot; - btrfs_debug(fs_info, "send_utimes %llu", ino); - - p = fs_path_alloc(); - if (!p) - return -ENOMEM; + p = get_path_for_command(sctx, ino, gen); + if (IS_ERR(p)) + return PTR_ERR(p); path = alloc_path_for_send(); if (!path) { @@ -2540,50 +2708,103 @@ static int send_utimes(struct send_ctx *sctx, u64 ino, u64 gen) if (ret < 0) goto out; - ret = get_cur_path(sctx, ino, gen, p); - if (ret < 0) - goto out; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_ATIME, eb, &ii->atime); TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_MTIME, eb, &ii->mtime); TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_CTIME, eb, &ii->ctime); - /* TODO Add otime support when the otime patches get into upstream */ + if (sctx->proto >= 2) + TLV_PUT_BTRFS_TIMESPEC(sctx, BTRFS_SEND_A_OTIME, eb, &ii->otime); ret = send_cmd(sctx); tlv_put_failure: out: - fs_path_free(p); - btrfs_free_path(path); + free_path_for_command(sctx, p); return ret; } /* + * If the cache is full, we can't remove entries from it and do a call to + * send_utimes() for each respective inode, because we might be finishing + * processing an inode that is a directory and it just got renamed, and existing + * entries in the cache may refer to inodes that have the directory in their + * full path - in which case we would generate outdated paths (pre-rename) + * for the inodes that the cache entries point to. Instead of pruning the + * cache when inserting, do it after we finish processing each inode at + * finish_inode_if_needed(). + */ +static int cache_dir_utimes(struct send_ctx *sctx, u64 dir, u64 gen) +{ + struct btrfs_lru_cache_entry *entry; + int ret; + + entry = btrfs_lru_cache_lookup(&sctx->dir_utimes_cache, dir, gen); + if (entry != NULL) + return 0; + + /* Caching is optional, don't fail if we can't allocate memory. */ + entry = kmalloc(sizeof(*entry), GFP_KERNEL); + if (!entry) + return send_utimes(sctx, dir, gen); + + entry->key = dir; + entry->gen = gen; + + ret = btrfs_lru_cache_store(&sctx->dir_utimes_cache, entry, GFP_KERNEL); + ASSERT(ret != -EEXIST); + if (ret) { + kfree(entry); + return send_utimes(sctx, dir, gen); + } + + return 0; +} + +static int trim_dir_utimes_cache(struct send_ctx *sctx) +{ + while (sctx->dir_utimes_cache.size > SEND_MAX_DIR_UTIMES_CACHE_SIZE) { + struct btrfs_lru_cache_entry *lru; + int ret; + + lru = btrfs_lru_cache_lru_entry(&sctx->dir_utimes_cache); + ASSERT(lru != NULL); + + ret = send_utimes(sctx, lru->key, lru->gen); + if (ret) + return ret; + + btrfs_lru_cache_remove(&sctx->dir_utimes_cache, lru); + } + + return 0; +} + +/* * Sends a BTRFS_SEND_C_MKXXX or SYMLINK command to user space. We don't have * a valid path yet because we did not process the refs yet. So, the inode * is created as orphan. */ static int send_create_inode(struct send_ctx *sctx, u64 ino) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret = 0; struct fs_path *p; int cmd; + struct btrfs_inode_info info; u64 gen; u64 mode; u64 rdev; - btrfs_debug(fs_info, "send_create_inode %llu", ino); - p = fs_path_alloc(); if (!p) return -ENOMEM; if (ino != sctx->cur_ino) { - ret = get_inode_info(sctx->send_root, ino, NULL, &gen, &mode, - NULL, NULL, &rdev); + ret = get_inode_info(sctx->send_root, ino, &info); if (ret < 0) goto out; + gen = info.gen; + mode = info.mode; + rdev = info.rdev; } else { gen = sctx->cur_inode_gen; mode = sctx->cur_inode_mode; @@ -2643,6 +2864,23 @@ out: return ret; } +static void cache_dir_created(struct send_ctx *sctx, u64 dir) +{ + struct btrfs_lru_cache_entry *entry; + int ret; + + /* Caching is optional, ignore any failures. */ + entry = kmalloc(sizeof(*entry), GFP_KERNEL); + if (!entry) + return; + + entry->key = dir; + entry->gen = 0; + ret = btrfs_lru_cache_store(&sctx->dir_created_cache, entry, GFP_KERNEL); + if (ret < 0) + kfree(entry); +} + /* * We need some special handling for inodes that get processed before the parent * directory got created. See process_recorded_refs for details. @@ -2651,62 +2889,47 @@ out: static int did_create_dir(struct send_ctx *sctx, u64 dir) { int ret = 0; - struct btrfs_path *path = NULL; + int iter_ret = 0; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; struct btrfs_key found_key; struct btrfs_key di_key; - struct extent_buffer *eb; struct btrfs_dir_item *di; - int slot; + + if (btrfs_lru_cache_lookup(&sctx->dir_created_cache, dir, 0)) + return 1; path = alloc_path_for_send(); - if (!path) { - ret = -ENOMEM; - goto out; - } + if (!path) + return -ENOMEM; key.objectid = dir; key.type = BTRFS_DIR_INDEX_KEY; key.offset = 0; - ret = btrfs_search_slot(NULL, sctx->send_root, &key, path, 0, 0); - if (ret < 0) - goto out; - while (1) { - eb = path->nodes[0]; - slot = path->slots[0]; - if (slot >= btrfs_header_nritems(eb)) { - ret = btrfs_next_leaf(sctx->send_root, path); - if (ret < 0) { - goto out; - } else if (ret > 0) { - ret = 0; - break; - } - continue; - } + btrfs_for_each_slot(sctx->send_root, &key, &found_key, path, iter_ret) { + struct extent_buffer *eb = path->nodes[0]; - btrfs_item_key_to_cpu(eb, &found_key, slot); if (found_key.objectid != key.objectid || found_key.type != key.type) { ret = 0; - goto out; + break; } - di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item); + di = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dir_item); btrfs_dir_item_key_to_cpu(eb, di, &di_key); if (di_key.type != BTRFS_ROOT_ITEM_KEY && di_key.objectid < sctx->send_progress) { ret = 1; - goto out; + cache_dir_created(sctx, dir); + break; } - - path->slots[0]++; } + /* Catch error found during iteration */ + if (iter_ret < 0) + ret = iter_ret; -out: - btrfs_free_path(path); return ret; } @@ -2723,18 +2946,16 @@ static int send_create_inode_if_needed(struct send_ctx *sctx) if (S_ISDIR(sctx->cur_inode_mode)) { ret = did_create_dir(sctx, sctx->cur_ino); if (ret < 0) - goto out; - if (ret) { - ret = 0; - goto out; - } + return ret; + else if (ret > 0) + return 0; } ret = send_create_inode(sctx, sctx->cur_ino); - if (ret < 0) - goto out; -out: + if (ret == 0 && S_ISDIR(sctx->cur_inode_mode)) + cache_dir_created(sctx, sctx->cur_ino); + return ret; } @@ -2745,48 +2966,50 @@ struct recorded_ref { u64 dir; u64 dir_gen; int name_len; + struct rb_node node; + struct rb_root *root; }; -static void set_ref_path(struct recorded_ref *ref, struct fs_path *path) +static struct recorded_ref *recorded_ref_alloc(void) { - ref->full_path = path; - ref->name = (char *)kbasename(ref->full_path->start); - ref->name_len = ref->full_path->end - ref->name; + struct recorded_ref *ref; + + ref = kzalloc(sizeof(*ref), GFP_KERNEL); + if (!ref) + return NULL; + RB_CLEAR_NODE(&ref->node); + INIT_LIST_HEAD(&ref->list); + return ref; } -/* - * We need to process new refs before deleted refs, but compare_tree gives us - * everything mixed. So we first record all refs and later process them. - * This function is a helper to record one ref. - */ -static int __record_ref(struct list_head *head, u64 dir, - u64 dir_gen, struct fs_path *path) +static void recorded_ref_free(struct recorded_ref *ref) { - struct recorded_ref *ref; - - ref = kmalloc(sizeof(*ref), GFP_KERNEL); if (!ref) - return -ENOMEM; + return; + if (!RB_EMPTY_NODE(&ref->node)) + rb_erase(&ref->node, ref->root); + list_del(&ref->list); + fs_path_free(ref->full_path); + kfree(ref); +} - ref->dir = dir; - ref->dir_gen = dir_gen; - set_ref_path(ref, path); - list_add_tail(&ref->list, head); - return 0; +static void set_ref_path(struct recorded_ref *ref, struct fs_path *path) +{ + ref->full_path = path; + ref->name = (char *)kbasename(ref->full_path->start); + ref->name_len = ref->full_path->end - ref->name; } static int dup_ref(struct recorded_ref *ref, struct list_head *list) { struct recorded_ref *new; - new = kmalloc(sizeof(*ref), GFP_KERNEL); + new = recorded_ref_alloc(); if (!new) return -ENOMEM; new->dir = ref->dir; new->dir_gen = ref->dir_gen; - new->full_path = NULL; - INIT_LIST_HEAD(&new->list); list_add_tail(&new->list, list); return 0; } @@ -2796,10 +3019,8 @@ static void __free_recorded_refs(struct list_head *head) struct recorded_ref *cur; while (!list_empty(head)) { - cur = list_entry(head->next, struct recorded_ref, list); - fs_path_free(cur->full_path); - list_del(&cur->list); - kfree(cur); + cur = list_first_entry(head, struct recorded_ref, list); + recorded_ref_free(cur); } } @@ -2829,14 +3050,19 @@ static int orphanize_inode(struct send_ctx *sctx, u64 ino, u64 gen, goto out; ret = send_rename(sctx, path, orphan); + if (ret < 0) + goto out; + + if (ino == sctx->cur_ino && gen == sctx->cur_inode_gen) + ret = fs_path_copy(&sctx->cur_inode_path, orphan); out: fs_path_free(orphan); return ret; } -static struct orphan_dir_info * -add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) +static struct orphan_dir_info *add_orphan_dir_info(struct send_ctx *sctx, + u64 dir_ino, u64 dir_gen) { struct rb_node **p = &sctx->orphan_dirs.rb_node; struct rb_node *parent = NULL; @@ -2845,29 +3071,33 @@ add_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) while (*p) { parent = *p; entry = rb_entry(parent, struct orphan_dir_info, node); - if (dir_ino < entry->ino) { + if (dir_ino < entry->ino) p = &(*p)->rb_left; - } else if (dir_ino > entry->ino) { + else if (dir_ino > entry->ino) p = &(*p)->rb_right; - } else { + else if (dir_gen < entry->gen) + p = &(*p)->rb_left; + else if (dir_gen > entry->gen) + p = &(*p)->rb_right; + else return entry; - } } odi = kmalloc(sizeof(*odi), GFP_KERNEL); if (!odi) return ERR_PTR(-ENOMEM); odi->ino = dir_ino; - odi->gen = 0; + odi->gen = dir_gen; odi->last_dir_index_offset = 0; + odi->dir_high_seq_ino = 0; rb_link_node(&odi->node, parent, p); rb_insert_color(&odi->node, &sctx->orphan_dirs); return odi; } -static struct orphan_dir_info * -get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) +static struct orphan_dir_info *get_orphan_dir_info(struct send_ctx *sctx, + u64 dir_ino, u64 gen) { struct rb_node *n = sctx->orphan_dirs.rb_node; struct orphan_dir_info *entry; @@ -2878,15 +3108,19 @@ get_orphan_dir_info(struct send_ctx *sctx, u64 dir_ino) n = n->rb_left; else if (dir_ino > entry->ino) n = n->rb_right; + else if (gen < entry->gen) + n = n->rb_left; + else if (gen > entry->gen) + n = n->rb_right; else return entry; } return NULL; } -static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino) +static int is_waiting_for_rm(struct send_ctx *sctx, u64 dir_ino, u64 gen) { - struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino); + struct orphan_dir_info *odi = get_orphan_dir_info(sctx, dir_ino, gen); return odi != NULL; } @@ -2905,10 +3139,10 @@ static void free_orphan_dir_info(struct send_ctx *sctx, * We check this by iterating all dir items and checking if the inode behind * the dir item was already processed. */ -static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, - u64 send_progress) +static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen) { int ret = 0; + int iter_ret = 0; struct btrfs_root *root = sctx->parent_root; struct btrfs_path *path; struct btrfs_key key; @@ -2916,6 +3150,8 @@ static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, struct btrfs_key loc; struct btrfs_dir_item *di; struct orphan_dir_info *odi = NULL; + u64 dir_high_seq_ino = 0; + u64 last_dir_index_offset = 0; /* * Don't try to rmdir the top/root subvolume dir. @@ -2923,35 +3159,66 @@ static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, if (dir == BTRFS_FIRST_FREE_OBJECTID) return 0; + odi = get_orphan_dir_info(sctx, dir, dir_gen); + if (odi && sctx->cur_ino < odi->dir_high_seq_ino) + return 0; + path = alloc_path_for_send(); if (!path) return -ENOMEM; - key.objectid = dir; - key.type = BTRFS_DIR_INDEX_KEY; - key.offset = 0; + if (!odi) { + /* + * Find the inode number associated with the last dir index + * entry. This is very likely the inode with the highest number + * of all inodes that have an entry in the directory. We can + * then use it to avoid future calls to can_rmdir(), when + * processing inodes with a lower number, from having to search + * the parent root b+tree for dir index keys. + */ + key.objectid = dir; + key.type = BTRFS_DIR_INDEX_KEY; + key.offset = (u64)-1; - odi = get_orphan_dir_info(sctx, dir); - if (odi) - key.offset = odi->last_dir_index_offset; + ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); + if (ret < 0) { + goto out; + } else if (ret > 0) { + /* Can't happen, the root is never empty. */ + ASSERT(path->slots[0] > 0); + if (WARN_ON(path->slots[0] == 0)) { + ret = -EUCLEAN; + goto out; + } + path->slots[0]--; + } - ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); - if (ret < 0) - goto out; + btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); + if (key.objectid != dir || key.type != BTRFS_DIR_INDEX_KEY) { + /* No index keys, dir can be removed. */ + ret = 1; + goto out; + } - while (1) { + di = btrfs_item_ptr(path->nodes[0], path->slots[0], + struct btrfs_dir_item); + btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); + dir_high_seq_ino = loc.objectid; + if (sctx->cur_ino < dir_high_seq_ino) { + ret = 0; + goto out; + } + + btrfs_release_path(path); + } + + key.objectid = dir; + key.type = BTRFS_DIR_INDEX_KEY; + key.offset = (odi ? odi->last_dir_index_offset : 0); + + btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { struct waiting_dir_move *dm; - if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { - ret = btrfs_next_leaf(root, path); - if (ret < 0) - goto out; - else if (ret > 0) - break; - continue; - } - btrfs_item_key_to_cpu(path->nodes[0], &found_key, - path->slots[0]); if (found_key.objectid != key.objectid || found_key.type != key.type) break; @@ -2960,33 +3227,25 @@ static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, struct btrfs_dir_item); btrfs_dir_item_key_to_cpu(path->nodes[0], di, &loc); + dir_high_seq_ino = max(dir_high_seq_ino, loc.objectid); + last_dir_index_offset = found_key.offset; + dm = get_waiting_dir_move(sctx, loc.objectid); if (dm) { - odi = add_orphan_dir_info(sctx, dir); - if (IS_ERR(odi)) { - ret = PTR_ERR(odi); - goto out; - } - odi->gen = dir_gen; - odi->last_dir_index_offset = found_key.offset; dm->rmdir_ino = dir; + dm->rmdir_gen = dir_gen; ret = 0; goto out; } - if (loc.objectid > send_progress) { - odi = add_orphan_dir_info(sctx, dir); - if (IS_ERR(odi)) { - ret = PTR_ERR(odi); - goto out; - } - odi->gen = dir_gen; - odi->last_dir_index_offset = found_key.offset; + if (loc.objectid > sctx->cur_ino) { ret = 0; goto out; } - - path->slots[0]++; + } + if (iter_ret < 0) { + ret = iter_ret; + goto out; } free_orphan_dir_info(sctx, odi); @@ -2994,7 +3253,22 @@ static int can_rmdir(struct send_ctx *sctx, u64 dir, u64 dir_gen, out: btrfs_free_path(path); - return ret; + + if (ret) + return ret; + + if (!odi) { + odi = add_orphan_dir_info(sctx, dir, dir_gen); + if (IS_ERR(odi)) + return PTR_ERR(odi); + + odi->gen = dir_gen; + } + + odi->last_dir_index_offset = last_dir_index_offset; + odi->dir_high_seq_ino = max(odi->dir_high_seq_ino, dir_high_seq_ino); + + return 0; } static int is_waiting_for_move(struct send_ctx *sctx, u64 ino) @@ -3015,6 +3289,7 @@ static int add_waiting_dir_move(struct send_ctx *sctx, u64 ino, bool orphanized) return -ENOMEM; dm->ino = ino; dm->rmdir_ino = 0; + dm->rmdir_gen = 0; dm->orphanized = orphanized; while (*p) { @@ -3160,7 +3435,7 @@ static int path_loop(struct send_ctx *sctx, struct fs_path *name, while (ino != BTRFS_FIRST_FREE_OBJECTID) { fs_path_reset(name); - if (is_waiting_for_rm(sctx, ino)) + if (is_waiting_for_rm(sctx, ino, gen)) break; if (is_waiting_for_move(sctx, ino)) { if (*ancestor_ino == 0) @@ -3200,6 +3475,7 @@ static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) u64 parent_ino, parent_gen; struct waiting_dir_move *dm = NULL; u64 rmdir_ino = 0; + u64 rmdir_gen; u64 ancestor; bool is_orphan; int ret; @@ -3214,6 +3490,7 @@ static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) dm = get_waiting_dir_move(sctx, pm->ino); ASSERT(dm); rmdir_ino = dm->rmdir_ino; + rmdir_gen = dm->rmdir_gen; is_orphan = dm->orphanized; free_waiting_dir_move(sctx, dm); @@ -3250,6 +3527,7 @@ static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) dm = get_waiting_dir_move(sctx, pm->ino); ASSERT(dm); dm->rmdir_ino = rmdir_ino; + dm->rmdir_gen = rmdir_gen; } goto out; } @@ -3268,14 +3546,14 @@ static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) struct orphan_dir_info *odi; u64 gen; - odi = get_orphan_dir_info(sctx, rmdir_ino); + odi = get_orphan_dir_info(sctx, rmdir_ino, rmdir_gen); if (!odi) { /* already deleted */ goto finish; } gen = odi->gen; - ret = can_rmdir(sctx, rmdir_ino, gen, sctx->cur_ino); + ret = can_rmdir(sctx, rmdir_ino, gen); if (ret < 0) goto out; if (!ret) @@ -3295,7 +3573,7 @@ static int apply_dir_move(struct send_ctx *sctx, struct pending_dir_move *pm) } finish: - ret = send_utimes(sctx, pm->ino, pm->gen); + ret = cache_dir_utimes(sctx, pm->ino, pm->gen); if (ret < 0) goto out; @@ -3307,8 +3585,7 @@ finish: /* * The parent inode might have been deleted in the send snapshot */ - ret = get_inode_info(sctx->send_root, cur->dir, NULL, - NULL, NULL, NULL, NULL, NULL); + ret = get_inode_info(sctx->send_root, cur->dir, NULL); if (ret == -ENOENT) { ret = 0; continue; @@ -3316,7 +3593,7 @@ finish: if (ret < 0) goto out; - ret = send_utimes(sctx, cur->dir, cur->dir_gen); + ret = cache_dir_utimes(sctx, cur->dir, cur->dir_gen); if (ret < 0) goto out; } @@ -3361,7 +3638,7 @@ static void tail_append_pending_moves(struct send_ctx *sctx, static int apply_children_dir_moves(struct send_ctx *sctx) { struct pending_dir_move *pm; - struct list_head stack; + LIST_HEAD(stack); u64 parent_ino = sctx->cur_ino; int ret = 0; @@ -3369,7 +3646,6 @@ static int apply_children_dir_moves(struct send_ctx *sctx) if (!pm) return 0; - INIT_LIST_HEAD(&stack); tail_append_pending_moves(sctx, pm, &stack); while (!list_empty(&stack)) { @@ -3433,8 +3709,7 @@ static int wait_for_dest_dir_move(struct send_ctx *sctx, struct recorded_ref *parent_ref, const bool is_orphan) { - struct btrfs_fs_info *fs_info = sctx->parent_root->fs_info; - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; struct btrfs_key di_key; struct btrfs_dir_item *di; @@ -3455,19 +3730,15 @@ static int wait_for_dest_dir_move(struct send_ctx *sctx, key.offset = btrfs_name_hash(parent_ref->name, parent_ref->name_len); ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); - if (ret < 0) { - goto out; - } else if (ret > 0) { - ret = 0; - goto out; - } + if (ret < 0) + return ret; + if (ret > 0) + return 0; - di = btrfs_match_dir_item_name(fs_info, path, parent_ref->name, + di = btrfs_match_dir_item_name(path, parent_ref->name, parent_ref->name_len); - if (!di) { - ret = 0; - goto out; - } + if (!di) + return 0; /* * di_key.objectid has the number of the inode that has a dentry in the * parent directory with the same name that sctx->cur_ino is being @@ -3477,28 +3748,22 @@ static int wait_for_dest_dir_move(struct send_ctx *sctx, * that it happens after that other inode is renamed. */ btrfs_dir_item_key_to_cpu(path->nodes[0], di, &di_key); - if (di_key.type != BTRFS_INODE_ITEM_KEY) { - ret = 0; - goto out; - } + if (di_key.type != BTRFS_INODE_ITEM_KEY) + return 0; - ret = get_inode_info(sctx->parent_root, di_key.objectid, NULL, - &left_gen, NULL, NULL, NULL, NULL); + ret = get_inode_gen(sctx->parent_root, di_key.objectid, &left_gen); if (ret < 0) - goto out; - ret = get_inode_info(sctx->send_root, di_key.objectid, NULL, - &right_gen, NULL, NULL, NULL, NULL); + return ret; + ret = get_inode_gen(sctx->send_root, di_key.objectid, &right_gen); if (ret < 0) { if (ret == -ENOENT) ret = 0; - goto out; + return ret; } /* Different inode, no need to delay the rename of sctx->cur_ino */ - if (right_gen != left_gen) { - ret = 0; - goto out; - } + if (right_gen != left_gen) + return 0; wdm = get_waiting_dir_move(sctx, di_key.objectid); if (wdm && !wdm->orphanized) { @@ -3512,8 +3777,6 @@ static int wait_for_dest_dir_move(struct send_ctx *sctx, if (!ret) ret = 1; } -out: - btrfs_free_path(path); return ret; } @@ -3550,7 +3813,7 @@ static int check_ino_in_path(struct btrfs_root *root, } /* - * Check if ino ino1 is an ancestor of inode ino2 in the given root for any + * Check if inode ino1 is an ancestor of inode ino2 in the given root for any * possible path (in case ino2 is not a directory and has multiple hard links). * Return 1 if true, 0 if false and < 0 on error. */ @@ -3562,7 +3825,8 @@ static int is_ancestor(struct btrfs_root *root, { bool free_fs_path = false; int ret = 0; - struct btrfs_path *path = NULL; + int iter_ret = 0; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; if (!fs_path) { @@ -3582,33 +3846,19 @@ static int is_ancestor(struct btrfs_root *root, key.type = BTRFS_INODE_REF_KEY; key.offset = 0; - ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); - if (ret < 0) - goto out; - - while (true) { + btrfs_for_each_slot(root, &key, &key, path, iter_ret) { struct extent_buffer *leaf = path->nodes[0]; int slot = path->slots[0]; u32 cur_offset = 0; u32 item_size; - if (slot >= btrfs_header_nritems(leaf)) { - ret = btrfs_next_leaf(root, path); - if (ret < 0) - goto out; - if (ret > 0) - break; - continue; - } - - btrfs_item_key_to_cpu(leaf, &key, slot); if (key.objectid != ino2) break; if (key.type != BTRFS_INODE_REF_KEY && key.type != BTRFS_INODE_EXTREF_KEY) break; - item_size = btrfs_item_size_nr(leaf, slot); + item_size = btrfs_item_size(leaf, slot); while (cur_offset < item_size) { u64 parent; u64 parent_gen; @@ -3630,8 +3880,7 @@ static int is_ancestor(struct btrfs_root *root, cur_offset = item_size; } - ret = get_inode_info(root, parent, NULL, &parent_gen, - NULL, NULL, NULL, NULL); + ret = get_inode_gen(root, parent, &parent_gen); if (ret < 0) goto out; ret = check_ino_in_path(root, ino1, ino1_gen, @@ -3639,11 +3888,12 @@ static int is_ancestor(struct btrfs_root *root, if (ret) goto out; } - path->slots[0]++; } ret = 0; - out: - btrfs_free_path(path); + if (iter_ret < 0) + ret = iter_ret; + +out: if (free_fs_path) fs_path_free(fs_path); return ret; @@ -3719,9 +3969,7 @@ static int wait_for_parent_move(struct send_ctx *sctx, memcmp(path_before->start, path_after->start, len1))) { u64 parent_ino_gen; - ret = get_inode_info(sctx->parent_root, ino, NULL, - &parent_ino_gen, NULL, NULL, NULL, - NULL); + ret = get_inode_gen(sctx->parent_root, ino, &parent_ino_gen); if (ret < 0) goto out; if (ino_gen == parent_ino_gen) { @@ -3783,6 +4031,130 @@ static int update_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) } /* + * When processing the new references for an inode we may orphanize an existing + * directory inode because its old name conflicts with one of the new references + * of the current inode. Later, when processing another new reference of our + * inode, we might need to orphanize another inode, but the path we have in the + * reference reflects the pre-orphanization name of the directory we previously + * orphanized. For example: + * + * parent snapshot looks like: + * + * . (ino 256) + * |----- f1 (ino 257) + * |----- f2 (ino 258) + * |----- d1/ (ino 259) + * |----- d2/ (ino 260) + * + * send snapshot looks like: + * + * . (ino 256) + * |----- d1 (ino 258) + * |----- f2/ (ino 259) + * |----- f2_link/ (ino 260) + * | |----- f1 (ino 257) + * | + * |----- d2 (ino 258) + * + * When processing inode 257 we compute the name for inode 259 as "d1", and we + * cache it in the name cache. Later when we start processing inode 258, when + * collecting all its new references we set a full path of "d1/d2" for its new + * reference with name "d2". When we start processing the new references we + * start by processing the new reference with name "d1", and this results in + * orphanizing inode 259, since its old reference causes a conflict. Then we + * move on the next new reference, with name "d2", and we find out we must + * orphanize inode 260, as its old reference conflicts with ours - but for the + * orphanization we use a source path corresponding to the path we stored in the + * new reference, which is "d1/d2" and not "o259-6-0/d2" - this makes the + * receiver fail since the path component "d1/" no longer exists, it was renamed + * to "o259-6-0/" when processing the previous new reference. So in this case we + * must recompute the path in the new reference and use it for the new + * orphanization operation. + */ +static int refresh_ref_path(struct send_ctx *sctx, struct recorded_ref *ref) +{ + char AUTO_KFREE(name); + int ret; + + name = kmemdup(ref->name, ref->name_len, GFP_KERNEL); + if (!name) + return -ENOMEM; + + fs_path_reset(ref->full_path); + ret = get_cur_path(sctx, ref->dir, ref->dir_gen, ref->full_path); + if (ret < 0) + return ret; + + ret = fs_path_add(ref->full_path, name, ref->name_len); + if (ret < 0) + return ret; + + /* Update the reference's base name pointer. */ + set_ref_path(ref, ref->full_path); + + return 0; +} + +static int rbtree_check_dir_ref_comp(const void *k, const struct rb_node *node) +{ + const struct recorded_ref *data = k; + const struct recorded_ref *ref = rb_entry(node, struct recorded_ref, node); + + if (data->dir > ref->dir) + return 1; + if (data->dir < ref->dir) + return -1; + if (data->dir_gen > ref->dir_gen) + return 1; + if (data->dir_gen < ref->dir_gen) + return -1; + return 0; +} + +static bool rbtree_check_dir_ref_less(struct rb_node *node, const struct rb_node *parent) +{ + const struct recorded_ref *entry = rb_entry(node, struct recorded_ref, node); + + return rbtree_check_dir_ref_comp(entry, parent) < 0; +} + +static int record_check_dir_ref_in_tree(struct rb_root *root, + struct recorded_ref *ref, struct list_head *list) +{ + struct recorded_ref *tmp_ref; + int ret; + + if (rb_find(ref, root, rbtree_check_dir_ref_comp)) + return 0; + + ret = dup_ref(ref, list); + if (ret < 0) + return ret; + + tmp_ref = list_last_entry(list, struct recorded_ref, list); + rb_add(&tmp_ref->node, root, rbtree_check_dir_ref_less); + tmp_ref->root = root; + return 0; +} + +static int rename_current_inode(struct send_ctx *sctx, + struct fs_path *current_path, + struct fs_path *new_path) +{ + int ret; + + ret = send_rename(sctx, current_path, new_path); + if (ret < 0) + return ret; + + ret = fs_path_copy(&sctx->cur_inode_path, new_path); + if (ret < 0) + return ret; + + return fs_path_copy(current_path, new_path); +} + +/* * This does all the move/link/unlink/rmdir magic. */ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) @@ -3791,26 +4163,29 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) int ret = 0; struct recorded_ref *cur; struct recorded_ref *cur2; - struct list_head check_dirs; + LIST_HEAD(check_dirs); + struct rb_root rbtree_check_dirs = RB_ROOT; struct fs_path *valid_path = NULL; u64 ow_inode = 0; u64 ow_gen; u64 ow_mode; - int did_overwrite = 0; - int is_orphan = 0; - u64 last_dir_ino_rm = 0; + bool did_overwrite = false; + bool is_orphan = false; bool can_rename = true; bool orphanized_dir = false; bool orphanized_ancestor = false; - btrfs_debug(fs_info, "process_recorded_refs %llu", sctx->cur_ino); - /* * This should never happen as the root dir always has the same ref * which is always '..' */ - BUG_ON(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID); - INIT_LIST_HEAD(&check_dirs); + if (unlikely(sctx->cur_ino <= BTRFS_FIRST_FREE_OBJECTID)) { + btrfs_err(fs_info, + "send: unexpected inode %llu in process_recorded_refs()", + sctx->cur_ino); + ret = -EINVAL; + goto out; + } valid_path = fs_path_alloc(); if (!valid_path) { @@ -3835,14 +4210,14 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) if (ret < 0) goto out; if (ret) - did_overwrite = 1; + did_overwrite = true; } if (sctx->cur_inode_new || did_overwrite) { ret = gen_unique_name(sctx, sctx->cur_ino, sctx->cur_inode_gen, valid_path); if (ret < 0) goto out; - is_orphan = 1; + is_orphan = true; } else { ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, valid_path); @@ -3850,52 +4225,56 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) goto out; } + /* + * Before doing any rename and link operations, do a first pass on the + * new references to orphanize any unprocessed inodes that may have a + * reference that conflicts with one of the new references of the current + * inode. This needs to happen first because a new reference may conflict + * with the old reference of a parent directory, so we must make sure + * that the path used for link and rename commands don't use an + * orphanized name when an ancestor was not yet orphanized. + * + * Example: + * + * Parent snapshot: + * + * . (ino 256) + * |----- testdir/ (ino 259) + * | |----- a (ino 257) + * | + * |----- b (ino 258) + * + * Send snapshot: + * + * . (ino 256) + * |----- testdir_2/ (ino 259) + * | |----- a (ino 260) + * | + * |----- testdir (ino 257) + * |----- b (ino 257) + * |----- b2 (ino 258) + * + * Processing the new reference for inode 257 with name "b" may happen + * before processing the new reference with name "testdir". If so, we + * must make sure that by the time we send a link command to create the + * hard link "b", inode 259 was already orphanized, since the generated + * path in "valid_path" already contains the orphanized name for 259. + * We are processing inode 257, so only later when processing 259 we do + * the rename operation to change its temporary (orphanized) name to + * "testdir_2". + */ list_for_each_entry(cur, &sctx->new_refs, list) { - /* - * We may have refs where the parent directory does not exist - * yet. This happens if the parent directories inum is higher - * than the current inum. To handle this case, we create the - * parent directory out of order. But we need to check if this - * did already happen before due to other refs in the same dir. - */ - ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); + ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen, NULL, NULL); if (ret < 0) goto out; - if (ret == inode_state_will_create) { - ret = 0; - /* - * First check if any of the current inodes refs did - * already create the dir. - */ - list_for_each_entry(cur2, &sctx->new_refs, list) { - if (cur == cur2) - break; - if (cur2->dir == cur->dir) { - ret = 1; - break; - } - } - - /* - * If that did not happen, check if a previous inode - * did already create the dir. - */ - if (!ret) - ret = did_create_dir(sctx, cur->dir); - if (ret < 0) - goto out; - if (!ret) { - ret = send_create_inode(sctx, cur->dir); - if (ret < 0) - goto out; - } - } + if (ret == inode_state_will_create) + continue; /* - * Check if this new ref would overwrite the first ref of - * another unprocessed inode. If yes, orphanize the - * overwritten inode. If we find an overwritten ref that is - * not the first ref, simply unlink it. + * Check if this new ref would overwrite the first ref of another + * unprocessed inode. If yes, orphanize the overwritten inode. + * If we find an overwritten ref that is not the first ref, + * simply unlink it. */ ret = will_overwrite_ref(sctx, cur->dir, cur->dir_gen, cur->name, cur->name_len, @@ -3912,6 +4291,12 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) struct name_cache_entry *nce; struct waiting_dir_move *wdm; + if (orphanized_dir) { + ret = refresh_ref_path(sctx, cur); + if (ret < 0) + goto out; + } + ret = orphanize_inode(sctx, ow_inode, ow_gen, cur->full_path); if (ret < 0) @@ -3925,12 +4310,9 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) * the source path when performing its rename * operation. */ - if (is_waiting_for_move(sctx, ow_inode)) { - wdm = get_waiting_dir_move(sctx, - ow_inode); - ASSERT(wdm); + wdm = get_waiting_dir_move(sctx, ow_inode); + if (wdm) wdm->orphanized = true; - } /* * Make sure we clear our orphanized inode's @@ -3943,10 +4325,9 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) * and get instead the orphan name. */ nce = name_cache_search(sctx, ow_inode, ow_gen); - if (nce) { - name_cache_delete(sctx, nce); - kfree(nce); - } + if (nce) + btrfs_lru_cache_remove(&sctx->name_cache, + &nce->entry); /* * ow_inode might currently be an ancestor of @@ -3961,6 +4342,7 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) if (ret > 0) { orphanized_ancestor = true; fs_path_reset(valid_path); + fs_path_reset(&sctx->cur_inode_path); ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, valid_path); @@ -3968,12 +4350,67 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) if (ret < 0) goto out; } else { + /* + * If we previously orphanized a directory that + * collided with a new reference that we already + * processed, recompute the current path because + * that directory may be part of the path. + */ + if (orphanized_dir) { + ret = refresh_ref_path(sctx, cur); + if (ret < 0) + goto out; + } ret = send_unlink(sctx, cur->full_path); if (ret < 0) goto out; } } + } + + list_for_each_entry(cur, &sctx->new_refs, list) { + /* + * We may have refs where the parent directory does not exist + * yet. This happens if the parent directories inum is higher + * than the current inum. To handle this case, we create the + * parent directory out of order. But we need to check if this + * did already happen before due to other refs in the same dir. + */ + ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen, NULL, NULL); + if (ret < 0) + goto out; + if (ret == inode_state_will_create) { + ret = 0; + /* + * First check if any of the current inodes refs did + * already create the dir. + */ + list_for_each_entry(cur2, &sctx->new_refs, list) { + if (cur == cur2) + break; + if (cur2->dir == cur->dir) { + ret = 1; + break; + } + } + + /* + * If that did not happen, check if a previous inode + * did already create the dir. + */ + if (!ret) + ret = did_create_dir(sctx, cur->dir); + if (ret < 0) + goto out; + if (!ret) { + ret = send_create_inode(sctx, cur->dir); + if (ret < 0) + goto out; + cache_dir_created(sctx, cur->dir); + } + } + if (S_ISDIR(sctx->cur_inode_mode) && sctx->parent_root) { ret = wait_for_dest_dir_move(sctx, cur, is_orphan); if (ret < 0) @@ -4001,13 +4438,10 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) * it depending on the inode mode. */ if (is_orphan && can_rename) { - ret = send_rename(sctx, valid_path, cur->full_path); - if (ret < 0) - goto out; - is_orphan = 0; - ret = fs_path_copy(valid_path, cur->full_path); + ret = rename_current_inode(sctx, valid_path, cur->full_path); if (ret < 0) goto out; + is_orphan = false; } else if (can_rename) { if (S_ISDIR(sctx->cur_inode_mode)) { /* @@ -4015,10 +4449,7 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) * dirs, we always have one new and one deleted * ref. The deleted ref is ignored later. */ - ret = send_rename(sctx, valid_path, - cur->full_path); - if (!ret) - ret = fs_path_copy(valid_path, + ret = rename_current_inode(sctx, valid_path, cur->full_path); if (ret < 0) goto out; @@ -4041,7 +4472,7 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) goto out; } } - ret = dup_ref(cur, &check_dirs); + ret = record_check_dir_ref_in_tree(&rbtree_check_dirs, cur, &check_dirs); if (ret < 0) goto out; } @@ -4053,8 +4484,7 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) * later, we do this check again and rmdir it then if possible. * See the use of check_dirs for more details. */ - ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen, - sctx->cur_ino); + ret = can_rmdir(sctx, sctx->cur_ino, sctx->cur_inode_gen); if (ret < 0) goto out; if (ret) { @@ -4066,11 +4496,11 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) sctx->cur_inode_gen, valid_path); if (ret < 0) goto out; - is_orphan = 1; + is_orphan = true; } list_for_each_entry(cur, &sctx->deleted_refs, list) { - ret = dup_ref(cur, &check_dirs); + ret = record_check_dir_ref_in_tree(&rbtree_check_dirs, cur, &check_dirs); if (ret < 0) goto out; } @@ -4079,9 +4509,8 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) /* * We have a moved dir. Add the old parent to check_dirs */ - cur = list_entry(sctx->deleted_refs.next, struct recorded_ref, - list); - ret = dup_ref(cur, &check_dirs); + cur = list_first_entry(&sctx->deleted_refs, struct recorded_ref, list); + ret = record_check_dir_ref_in_tree(&rbtree_check_dirs, cur, &check_dirs); if (ret < 0) goto out; } else if (!S_ISDIR(sctx->cur_inode_mode)) { @@ -4112,8 +4541,10 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) ret = send_unlink(sctx, cur->full_path); if (ret < 0) goto out; + if (is_current_inode_path(sctx, cur->full_path)) + fs_path_reset(&sctx->cur_inode_path); } - ret = dup_ref(cur, &check_dirs); + ret = record_check_dir_ref_in_tree(&rbtree_check_dirs, cur, &check_dirs); if (ret < 0) goto out; } @@ -4147,20 +4578,17 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) if (cur->dir > sctx->cur_ino) continue; - ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen); + ret = get_cur_inode_state(sctx, cur->dir, cur->dir_gen, NULL, NULL); if (ret < 0) goto out; if (ret == inode_state_did_create || ret == inode_state_no_change) { - /* TODO delayed utimes */ - ret = send_utimes(sctx, cur->dir, cur->dir_gen); + ret = cache_dir_utimes(sctx, cur->dir, cur->dir_gen); if (ret < 0) goto out; - } else if (ret == inode_state_did_delete && - cur->dir != last_dir_ino_rm) { - ret = can_rmdir(sctx, cur->dir, cur->dir_gen, - sctx->cur_ino); + } else if (ret == inode_state_did_delete) { + ret = can_rmdir(sctx, cur->dir, cur->dir_gen); if (ret < 0) goto out; if (ret) { @@ -4171,7 +4599,6 @@ static int process_recorded_refs(struct send_ctx *sctx, int *pending_move) ret = send_rmdir(sctx, valid_path); if (ret < 0) goto out; - last_dir_ino_rm = cur->dir; } } } @@ -4185,204 +4612,171 @@ out: return ret; } -static int record_ref(struct btrfs_root *root, u64 dir, struct fs_path *name, - void *ctx, struct list_head *refs) +static int rbtree_ref_comp(const void *k, const struct rb_node *node) { - int ret = 0; - struct send_ctx *sctx = ctx; - struct fs_path *p; - u64 gen; + const struct recorded_ref *data = k; + const struct recorded_ref *ref = rb_entry(node, struct recorded_ref, node); - p = fs_path_alloc(); - if (!p) - return -ENOMEM; - - ret = get_inode_info(root, dir, NULL, &gen, NULL, NULL, - NULL, NULL); - if (ret < 0) - goto out; - - ret = get_cur_path(sctx, dir, gen, p); - if (ret < 0) - goto out; - ret = fs_path_add_path(p, name); - if (ret < 0) - goto out; - - ret = __record_ref(refs, dir, gen, p); - -out: - if (ret) - fs_path_free(p); - return ret; + if (data->dir > ref->dir) + return 1; + if (data->dir < ref->dir) + return -1; + if (data->dir_gen > ref->dir_gen) + return 1; + if (data->dir_gen < ref->dir_gen) + return -1; + if (data->name_len > ref->name_len) + return 1; + if (data->name_len < ref->name_len) + return -1; + return strcmp(data->name, ref->name); } -static int __record_new_ref(int num, u64 dir, int index, - struct fs_path *name, - void *ctx) +static bool rbtree_ref_less(struct rb_node *node, const struct rb_node *parent) { - struct send_ctx *sctx = ctx; - return record_ref(sctx->send_root, dir, name, ctx, &sctx->new_refs); -} - + const struct recorded_ref *entry = rb_entry(node, struct recorded_ref, node); -static int __record_deleted_ref(int num, u64 dir, int index, - struct fs_path *name, - void *ctx) -{ - struct send_ctx *sctx = ctx; - return record_ref(sctx->parent_root, dir, name, ctx, - &sctx->deleted_refs); + return rbtree_ref_comp(entry, parent) < 0; } -static int record_new_ref(struct send_ctx *sctx) +static int record_ref_in_tree(struct rb_root *root, struct list_head *refs, + struct fs_path *name, u64 dir, u64 dir_gen, + struct send_ctx *sctx) { - int ret; + int ret = 0; + struct fs_path *path = NULL; + struct recorded_ref *ref = NULL; - ret = iterate_inode_ref(sctx->send_root, sctx->left_path, - sctx->cmp_key, 0, __record_new_ref, sctx); - if (ret < 0) + path = fs_path_alloc(); + if (!path) { + ret = -ENOMEM; goto out; - ret = 0; - -out: - return ret; -} + } -static int record_deleted_ref(struct send_ctx *sctx) -{ - int ret; + ref = recorded_ref_alloc(); + if (!ref) { + ret = -ENOMEM; + goto out; + } - ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, - sctx->cmp_key, 0, __record_deleted_ref, sctx); + ret = get_cur_path(sctx, dir, dir_gen, path); + if (ret < 0) + goto out; + ret = fs_path_add_path(path, name); if (ret < 0) goto out; - ret = 0; + ref->dir = dir; + ref->dir_gen = dir_gen; + set_ref_path(ref, path); + list_add_tail(&ref->list, refs); + rb_add(&ref->node, root, rbtree_ref_less); + ref->root = root; out: + if (ret) { + if (path && (!ref || !ref->full_path)) + fs_path_free(path); + recorded_ref_free(ref); + } return ret; } -struct find_ref_ctx { - u64 dir; - u64 dir_gen; - struct btrfs_root *root; - struct fs_path *name; - int found_idx; -}; - -static int __find_iref(int num, u64 dir, int index, - struct fs_path *name, - void *ctx_) +static int record_new_ref_if_needed(u64 dir, struct fs_path *name, void *ctx) { - struct find_ref_ctx *ctx = ctx_; - u64 dir_gen; int ret; + struct send_ctx *sctx = ctx; + struct rb_node *node = NULL; + struct recorded_ref data; + struct recorded_ref *ref; + u64 dir_gen; - if (dir == ctx->dir && fs_path_len(name) == fs_path_len(ctx->name) && - strncmp(name->start, ctx->name->start, fs_path_len(name)) == 0) { - /* - * To avoid doing extra lookups we'll only do this if everything - * else matches. - */ - ret = get_inode_info(ctx->root, dir, NULL, &dir_gen, NULL, - NULL, NULL, NULL); - if (ret) - return ret; - if (dir_gen != ctx->dir_gen) - return 0; - ctx->found_idx = num; - return 1; + ret = get_inode_gen(sctx->send_root, dir, &dir_gen); + if (ret < 0) + return ret; + + data.dir = dir; + data.dir_gen = dir_gen; + set_ref_path(&data, name); + node = rb_find(&data, &sctx->rbtree_deleted_refs, rbtree_ref_comp); + if (node) { + ref = rb_entry(node, struct recorded_ref, node); + recorded_ref_free(ref); + } else { + ret = record_ref_in_tree(&sctx->rbtree_new_refs, + &sctx->new_refs, name, dir, dir_gen, + sctx); } - return 0; + + return ret; } -static int find_iref(struct btrfs_root *root, - struct btrfs_path *path, - struct btrfs_key *key, - u64 dir, u64 dir_gen, struct fs_path *name) +static int record_deleted_ref_if_needed(u64 dir, struct fs_path *name, void *ctx) { int ret; - struct find_ref_ctx ctx; - - ctx.dir = dir; - ctx.name = name; - ctx.dir_gen = dir_gen; - ctx.found_idx = -1; - ctx.root = root; + struct send_ctx *sctx = ctx; + struct rb_node *node = NULL; + struct recorded_ref data; + struct recorded_ref *ref; + u64 dir_gen; - ret = iterate_inode_ref(root, path, key, 0, __find_iref, &ctx); + ret = get_inode_gen(sctx->parent_root, dir, &dir_gen); if (ret < 0) return ret; - if (ctx.found_idx == -1) - return -ENOENT; + data.dir = dir; + data.dir_gen = dir_gen; + set_ref_path(&data, name); + node = rb_find(&data, &sctx->rbtree_new_refs, rbtree_ref_comp); + if (node) { + ref = rb_entry(node, struct recorded_ref, node); + recorded_ref_free(ref); + } else { + ret = record_ref_in_tree(&sctx->rbtree_deleted_refs, + &sctx->deleted_refs, name, dir, + dir_gen, sctx); + } - return ctx.found_idx; + return ret; } -static int __record_changed_new_ref(int num, u64 dir, int index, - struct fs_path *name, - void *ctx) +static int record_new_ref(struct send_ctx *sctx) { - u64 dir_gen; int ret; - struct send_ctx *sctx = ctx; - ret = get_inode_info(sctx->send_root, dir, NULL, &dir_gen, NULL, - NULL, NULL, NULL); - if (ret) + ret = iterate_inode_ref(sctx->send_root, sctx->left_path, sctx->cmp_key, + false, record_new_ref_if_needed, sctx); + if (ret < 0) return ret; - ret = find_iref(sctx->parent_root, sctx->right_path, - sctx->cmp_key, dir, dir_gen, name); - if (ret == -ENOENT) - ret = __record_new_ref(num, dir, index, name, sctx); - else if (ret > 0) - ret = 0; - - return ret; + return 0; } -static int __record_changed_deleted_ref(int num, u64 dir, int index, - struct fs_path *name, - void *ctx) +static int record_deleted_ref(struct send_ctx *sctx) { - u64 dir_gen; int ret; - struct send_ctx *sctx = ctx; - ret = get_inode_info(sctx->parent_root, dir, NULL, &dir_gen, NULL, - NULL, NULL, NULL); - if (ret) + ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, sctx->cmp_key, + false, record_deleted_ref_if_needed, sctx); + if (ret < 0) return ret; - ret = find_iref(sctx->send_root, sctx->left_path, sctx->cmp_key, - dir, dir_gen, name); - if (ret == -ENOENT) - ret = __record_deleted_ref(num, dir, index, name, sctx); - else if (ret > 0) - ret = 0; - - return ret; + return 0; } static int record_changed_ref(struct send_ctx *sctx) { - int ret = 0; + int ret; - ret = iterate_inode_ref(sctx->send_root, sctx->left_path, - sctx->cmp_key, 0, __record_changed_new_ref, sctx); + ret = iterate_inode_ref(sctx->send_root, sctx->left_path, sctx->cmp_key, + false, record_new_ref_if_needed, sctx); if (ret < 0) - goto out; - ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, - sctx->cmp_key, 0, __record_changed_deleted_ref, sctx); + return ret; + ret = iterate_inode_ref(sctx->parent_root, sctx->right_path, sctx->cmp_key, + false, record_deleted_ref_if_needed, sctx); if (ret < 0) - goto out; - ret = 0; + return ret; -out: - return ret; + return 0; } /* @@ -4392,13 +4786,12 @@ out: static int process_all_refs(struct send_ctx *sctx, enum btrfs_compare_tree_result cmd) { - int ret; + int ret = 0; + int iter_ret = 0; struct btrfs_root *root; - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; struct btrfs_key found_key; - struct extent_buffer *eb; - int slot; iterate_inode_ref_t cb; int pending_move = 0; @@ -4408,49 +4801,33 @@ static int process_all_refs(struct send_ctx *sctx, if (cmd == BTRFS_COMPARE_TREE_NEW) { root = sctx->send_root; - cb = __record_new_ref; + cb = record_new_ref_if_needed; } else if (cmd == BTRFS_COMPARE_TREE_DELETED) { root = sctx->parent_root; - cb = __record_deleted_ref; + cb = record_deleted_ref_if_needed; } else { btrfs_err(sctx->send_root->fs_info, "Wrong command %d in process_all_refs", cmd); - ret = -EINVAL; - goto out; + return -EINVAL; } key.objectid = sctx->cmp_key->objectid; key.type = BTRFS_INODE_REF_KEY; key.offset = 0; - ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); - if (ret < 0) - goto out; - - while (1) { - eb = path->nodes[0]; - slot = path->slots[0]; - if (slot >= btrfs_header_nritems(eb)) { - ret = btrfs_next_leaf(root, path); - if (ret < 0) - goto out; - else if (ret > 0) - break; - continue; - } - - btrfs_item_key_to_cpu(eb, &found_key, slot); - + btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { if (found_key.objectid != key.objectid || (found_key.type != BTRFS_INODE_REF_KEY && found_key.type != BTRFS_INODE_EXTREF_KEY)) break; - ret = iterate_inode_ref(root, path, &found_key, 0, cb, sctx); + ret = iterate_inode_ref(root, path, &found_key, false, cb, sctx); if (ret < 0) - goto out; - - path->slots[0]++; + return ret; } + /* Catch error found during iteration */ + if (iter_ret < 0) + return iter_ret; + btrfs_release_path(path); /* @@ -4458,22 +4835,23 @@ static int process_all_refs(struct send_ctx *sctx, * re-creating this inode and will be rename'ing it into place once we * rename the parent directory. */ - ret = process_recorded_refs(sctx, &pending_move); -out: - btrfs_free_path(path); - return ret; + return process_recorded_refs(sctx, &pending_move); } static int send_set_xattr(struct send_ctx *sctx, - struct fs_path *path, const char *name, int name_len, const char *data, int data_len) { - int ret = 0; + struct fs_path *path; + int ret; + + path = get_cur_inode_path(sctx); + if (IS_ERR(path)) + return PTR_ERR(path); ret = begin_cmd(sctx, BTRFS_SEND_C_SET_XATTR); if (ret < 0) - goto out; + return ret; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); @@ -4482,7 +4860,6 @@ static int send_set_xattr(struct send_ctx *sctx, ret = send_cmd(sctx); tlv_put_failure: -out: return ret; } @@ -4490,11 +4867,11 @@ static int send_remove_xattr(struct send_ctx *sctx, struct fs_path *path, const char *name, int name_len) { - int ret = 0; + int ret; ret = begin_cmd(sctx, BTRFS_SEND_C_REMOVE_XATTR); if (ret < 0) - goto out; + return ret; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); TLV_PUT_STRING(sctx, BTRFS_SEND_A_XATTR_NAME, name, name_len); @@ -4502,23 +4879,19 @@ static int send_remove_xattr(struct send_ctx *sctx, ret = send_cmd(sctx); tlv_put_failure: -out: return ret; } static int __process_new_xattr(int num, struct btrfs_key *di_key, - const char *name, int name_len, - const char *data, int data_len, - u8 type, void *ctx) + const char *name, int name_len, const char *data, + int data_len, void *ctx) { - int ret; struct send_ctx *sctx = ctx; - struct fs_path *p; struct posix_acl_xattr_header dummy_acl; - p = fs_path_alloc(); - if (!p) - return -ENOMEM; + /* Capabilities are emitted by finish_inode_if_needed */ + if (!strncmp(name, XATTR_NAME_CAPS, name_len)) + return 0; /* * This hack is needed because empty acls are stored as zero byte @@ -4536,49 +4909,27 @@ static int __process_new_xattr(int num, struct btrfs_key *di_key, } } - ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); - if (ret < 0) - goto out; - - ret = send_set_xattr(sctx, p, name, name_len, data, data_len); - -out: - fs_path_free(p); - return ret; + return send_set_xattr(sctx, name, name_len, data, data_len); } static int __process_deleted_xattr(int num, struct btrfs_key *di_key, const char *name, int name_len, - const char *data, int data_len, - u8 type, void *ctx) + const char *data, int data_len, void *ctx) { - int ret; struct send_ctx *sctx = ctx; struct fs_path *p; - p = fs_path_alloc(); - if (!p) - return -ENOMEM; - - ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); - if (ret < 0) - goto out; + p = get_cur_inode_path(sctx); + if (IS_ERR(p)) + return PTR_ERR(p); - ret = send_remove_xattr(sctx, p, name, name_len); - -out: - fs_path_free(p); - return ret; + return send_remove_xattr(sctx, p, name, name_len); } static int process_new_xattr(struct send_ctx *sctx) { - int ret = 0; - - ret = iterate_dir_item(sctx->send_root, sctx->left_path, - __process_new_xattr, sctx); - - return ret; + return iterate_dir_item(sctx->send_root, sctx->left_path, + __process_new_xattr, sctx); } static int process_deleted_xattr(struct send_ctx *sctx) @@ -4593,12 +4944,11 @@ struct find_xattr_ctx { int found_idx; char *found_data; int found_data_len; + bool copy_data; }; -static int __find_xattr(int num, struct btrfs_key *di_key, - const char *name, int name_len, - const char *data, int data_len, - u8 type, void *vctx) +static int __find_xattr(int num, struct btrfs_key *di_key, const char *name, + int name_len, const char *data, int data_len, void *vctx) { struct find_xattr_ctx *ctx = vctx; @@ -4606,9 +4956,11 @@ static int __find_xattr(int num, struct btrfs_key *di_key, strncmp(name, ctx->name, name_len) == 0) { ctx->found_idx = num; ctx->found_data_len = data_len; - ctx->found_data = kmemdup(data, data_len, GFP_KERNEL); - if (!ctx->found_data) - return -ENOMEM; + if (ctx->copy_data) { + ctx->found_data = kmemdup(data, data_len, GFP_KERNEL); + if (!ctx->found_data) + return -ENOMEM; + } return 1; } return 0; @@ -4628,6 +4980,7 @@ static int find_xattr(struct btrfs_root *root, ctx.found_idx = -1; ctx.found_data = NULL; ctx.found_data_len = 0; + ctx.copy_data = (data != NULL); ret = iterate_dir_item(root, path, __find_xattr, &ctx); if (ret < 0) @@ -4639,7 +4992,7 @@ static int find_xattr(struct btrfs_root *root, *data = ctx.found_data; *data_len = ctx.found_data_len; } else { - kfree(ctx.found_data); + ASSERT(ctx.found_data == NULL); } return ctx.found_idx; } @@ -4648,37 +5001,36 @@ static int find_xattr(struct btrfs_root *root, static int __process_changed_new_xattr(int num, struct btrfs_key *di_key, const char *name, int name_len, const char *data, int data_len, - u8 type, void *ctx) + void *ctx) { int ret; struct send_ctx *sctx = ctx; - char *found_data = NULL; - int found_data_len = 0; + char AUTO_KFREE(found_data); + int found_data_len = 0; ret = find_xattr(sctx->parent_root, sctx->right_path, sctx->cmp_key, name, name_len, &found_data, &found_data_len); if (ret == -ENOENT) { ret = __process_new_xattr(num, di_key, name, name_len, data, - data_len, type, ctx); + data_len, ctx); } else if (ret >= 0) { if (data_len != found_data_len || memcmp(data, found_data, data_len)) { ret = __process_new_xattr(num, di_key, name, name_len, - data, data_len, type, ctx); + data, data_len, ctx); } else { ret = 0; } } - kfree(found_data); return ret; } static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, const char *name, int name_len, const char *data, int data_len, - u8 type, void *ctx) + void *ctx) { int ret; struct send_ctx *sctx = ctx; @@ -4687,7 +5039,7 @@ static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, name, name_len, NULL, NULL); if (ret == -ENOENT) ret = __process_deleted_xattr(num, di_key, name, name_len, data, - data_len, type, ctx); + data_len, ctx); else if (ret >= 0) ret = 0; @@ -4696,28 +5048,25 @@ static int __process_changed_deleted_xattr(int num, struct btrfs_key *di_key, static int process_changed_xattr(struct send_ctx *sctx) { - int ret = 0; + int ret; ret = iterate_dir_item(sctx->send_root, sctx->left_path, __process_changed_new_xattr, sctx); if (ret < 0) - goto out; - ret = iterate_dir_item(sctx->parent_root, sctx->right_path, - __process_changed_deleted_xattr, sctx); + return ret; -out: - return ret; + return iterate_dir_item(sctx->parent_root, sctx->right_path, + __process_changed_deleted_xattr, sctx); } static int process_all_new_xattrs(struct send_ctx *sctx) { - int ret; + int ret = 0; + int iter_ret = 0; struct btrfs_root *root; - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; struct btrfs_key found_key; - struct extent_buffer *eb; - int slot; path = alloc_path_for_send(); if (!path) @@ -4728,124 +5077,191 @@ static int process_all_new_xattrs(struct send_ctx *sctx) key.objectid = sctx->cmp_key->objectid; key.type = BTRFS_XATTR_ITEM_KEY; key.offset = 0; - ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); - if (ret < 0) - goto out; - - while (1) { - eb = path->nodes[0]; - slot = path->slots[0]; - if (slot >= btrfs_header_nritems(eb)) { - ret = btrfs_next_leaf(root, path); - if (ret < 0) { - goto out; - } else if (ret > 0) { - ret = 0; - break; - } - continue; - } - - btrfs_item_key_to_cpu(eb, &found_key, slot); + btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { if (found_key.objectid != key.objectid || found_key.type != key.type) { ret = 0; - goto out; + break; } ret = iterate_dir_item(root, path, __process_new_xattr, sctx); if (ret < 0) - goto out; - - path->slots[0]++; + break; } + /* Catch error found during iteration */ + if (iter_ret < 0) + ret = iter_ret; -out: - btrfs_free_path(path); return ret; } -static ssize_t fill_read_buf(struct send_ctx *sctx, u64 offset, u32 len) +static int send_verity(struct send_ctx *sctx, struct fs_path *path, + struct fsverity_descriptor *desc) { - struct btrfs_root *root = sctx->send_root; - struct btrfs_fs_info *fs_info = root->fs_info; - struct inode *inode; - struct page *page; - char *addr; - struct btrfs_key key; - pgoff_t index = offset >> PAGE_SHIFT; - pgoff_t last_index; - unsigned pg_offset = offset_in_page(offset); - ssize_t ret = 0; + int ret; - key.objectid = sctx->cur_ino; - key.type = BTRFS_INODE_ITEM_KEY; - key.offset = 0; + ret = begin_cmd(sctx, BTRFS_SEND_C_ENABLE_VERITY); + if (ret < 0) + return ret; + + TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); + TLV_PUT_U8(sctx, BTRFS_SEND_A_VERITY_ALGORITHM, + le8_to_cpu(desc->hash_algorithm)); + TLV_PUT_U32(sctx, BTRFS_SEND_A_VERITY_BLOCK_SIZE, + 1U << le8_to_cpu(desc->log_blocksize)); + TLV_PUT(sctx, BTRFS_SEND_A_VERITY_SALT_DATA, desc->salt, + le8_to_cpu(desc->salt_size)); + TLV_PUT(sctx, BTRFS_SEND_A_VERITY_SIG_DATA, desc->signature, + le32_to_cpu(desc->sig_size)); + + ret = send_cmd(sctx); + +tlv_put_failure: + return ret; +} + +static int process_verity(struct send_ctx *sctx) +{ + int ret = 0; + struct btrfs_inode *inode; + struct fs_path *p; - inode = btrfs_iget(fs_info->sb, &key, root, NULL); + inode = btrfs_iget(sctx->cur_ino, sctx->send_root); if (IS_ERR(inode)) return PTR_ERR(inode); - if (offset + len > i_size_read(inode)) { - if (offset > i_size_read(inode)) - len = 0; - else - len = offset - i_size_read(inode); + ret = btrfs_get_verity_descriptor(&inode->vfs_inode, NULL, 0); + if (ret < 0) + goto iput; + + if (unlikely(ret > FS_VERITY_MAX_DESCRIPTOR_SIZE)) { + ret = -EMSGSIZE; + goto iput; } - if (len == 0) - goto out; + if (!sctx->verity_descriptor) { + sctx->verity_descriptor = kvmalloc(FS_VERITY_MAX_DESCRIPTOR_SIZE, + GFP_KERNEL); + if (!sctx->verity_descriptor) { + ret = -ENOMEM; + goto iput; + } + } + + ret = btrfs_get_verity_descriptor(&inode->vfs_inode, sctx->verity_descriptor, ret); + if (ret < 0) + goto iput; + + p = get_cur_inode_path(sctx); + if (IS_ERR(p)) { + ret = PTR_ERR(p); + goto iput; + } + + ret = send_verity(sctx, p, sctx->verity_descriptor); +iput: + iput(&inode->vfs_inode); + return ret; +} + +static inline u64 max_send_read_size(const struct send_ctx *sctx) +{ + return sctx->send_max_size - SZ_16K; +} - last_index = (offset + len - 1) >> PAGE_SHIFT; +static int put_data_header(struct send_ctx *sctx, u32 len) +{ + if (WARN_ON_ONCE(sctx->put_data)) + return -EINVAL; + sctx->put_data = true; + if (sctx->proto >= 2) { + /* + * Since v2, the data attribute header doesn't include a length, + * it is implicitly to the end of the command. + */ + if (unlikely(sctx->send_max_size - sctx->send_size < sizeof(__le16) + len)) + return -EOVERFLOW; + put_unaligned_le16(BTRFS_SEND_A_DATA, sctx->send_buf + sctx->send_size); + sctx->send_size += sizeof(__le16); + } else { + struct btrfs_tlv_header *hdr; - /* initial readahead */ - memset(&sctx->ra, 0, sizeof(struct file_ra_state)); - file_ra_state_init(&sctx->ra, inode->i_mapping); + if (unlikely(sctx->send_max_size - sctx->send_size < sizeof(*hdr) + len)) + return -EOVERFLOW; + hdr = (struct btrfs_tlv_header *)(sctx->send_buf + sctx->send_size); + put_unaligned_le16(BTRFS_SEND_A_DATA, &hdr->tlv_type); + put_unaligned_le16(len, &hdr->tlv_len); + sctx->send_size += sizeof(*hdr); + } + return 0; +} - while (index <= last_index) { - unsigned cur_len = min_t(unsigned, len, - PAGE_SIZE - pg_offset); +static int put_file_data(struct send_ctx *sctx, u64 offset, u32 len) +{ + struct btrfs_root *root = sctx->send_root; + struct btrfs_fs_info *fs_info = root->fs_info; + u64 cur = offset; + const u64 end = offset + len; + const pgoff_t last_index = ((end - 1) >> PAGE_SHIFT); + struct address_space *mapping = sctx->cur_inode->i_mapping; + int ret; - page = find_lock_page(inode->i_mapping, index); - if (!page) { - page_cache_sync_readahead(inode->i_mapping, &sctx->ra, - NULL, index, last_index + 1 - index); + ret = put_data_header(sctx, len); + if (ret) + return ret; - page = find_or_create_page(inode->i_mapping, index, - GFP_KERNEL); - if (!page) { - ret = -ENOMEM; + while (cur < end) { + pgoff_t index = (cur >> PAGE_SHIFT); + unsigned int cur_len; + unsigned int pg_offset; + struct folio *folio; + + folio = filemap_lock_folio(mapping, index); + if (IS_ERR(folio)) { + page_cache_sync_readahead(mapping, + &sctx->ra, NULL, index, + last_index + 1 - index); + + folio = filemap_grab_folio(mapping, index); + if (IS_ERR(folio)) { + ret = PTR_ERR(folio); break; } } - - if (PageReadahead(page)) { - page_cache_async_readahead(inode->i_mapping, &sctx->ra, - NULL, page, index, last_index + 1 - index); - } - - if (!PageUptodate(page)) { - btrfs_readpage(NULL, page); - lock_page(page); - if (!PageUptodate(page)) { - unlock_page(page); - put_page(page); + pg_offset = offset_in_folio(folio, cur); + cur_len = min_t(unsigned int, end - cur, folio_size(folio) - pg_offset); + + if (folio_test_readahead(folio)) + page_cache_async_readahead(mapping, &sctx->ra, NULL, folio, + last_index + 1 - index); + + if (!folio_test_uptodate(folio)) { + btrfs_read_folio(NULL, folio); + folio_lock(folio); + if (unlikely(!folio_test_uptodate(folio))) { + folio_unlock(folio); + btrfs_err(fs_info, + "send: IO error at offset %llu for inode %llu root %llu", + folio_pos(folio), sctx->cur_ino, + btrfs_root_id(sctx->send_root)); + folio_put(folio); ret = -EIO; break; } + if (folio->mapping != mapping) { + folio_unlock(folio); + folio_put(folio); + continue; + } } - addr = kmap(page); - memcpy(sctx->read_buf + ret, addr + pg_offset, cur_len); - kunmap(page); - unlock_page(page); - put_page(page); - index++; - pg_offset = 0; - len -= cur_len; - ret += cur_len; + memcpy_from_folio(sctx->send_buf + sctx->send_size, folio, + pg_offset, cur_len); + folio_unlock(folio); + folio_put(folio); + cur += cur_len; + sctx->send_size += cur_len; } -out: - iput(inode); + return ret; } @@ -4855,44 +5271,27 @@ out: */ static int send_write(struct send_ctx *sctx, u64 offset, u32 len) { - struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; int ret = 0; struct fs_path *p; - ssize_t num_read = 0; - p = fs_path_alloc(); - if (!p) - return -ENOMEM; - - btrfs_debug(fs_info, "send_write offset=%llu, len=%d", offset, len); - - num_read = fill_read_buf(sctx, offset, len); - if (num_read <= 0) { - if (num_read < 0) - ret = num_read; - goto out; - } + p = get_cur_inode_path(sctx); + if (IS_ERR(p)) + return PTR_ERR(p); ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); if (ret < 0) - goto out; - - ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); - if (ret < 0) - goto out; + return ret; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); - TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, num_read); + ret = put_file_data(sctx, offset, len); + if (ret < 0) + return ret; ret = send_cmd(sctx); tlv_put_failure: -out: - fs_path_free(p); - if (ret < 0) - return ret; - return num_read; + return ret; } /* @@ -4904,12 +5303,12 @@ static int send_clone(struct send_ctx *sctx, { int ret = 0; struct fs_path *p; + struct fs_path *cur_inode_path; u64 gen; - btrfs_debug(sctx->send_root->fs_info, - "send_clone offset=%llu, len=%d, clone_root=%llu, clone_inode=%llu, clone_offset=%llu", - offset, len, clone_root->root->root_key.objectid, - clone_root->ino, clone_root->offset); + cur_inode_path = get_cur_inode_path(sctx); + if (IS_ERR(cur_inode_path)) + return PTR_ERR(cur_inode_path); p = fs_path_alloc(); if (!p) @@ -4919,17 +5318,12 @@ static int send_clone(struct send_ctx *sctx, if (ret < 0) goto out; - ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); - if (ret < 0) - goto out; - TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_LEN, len); - TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); + TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, cur_inode_path); if (clone_root->root == sctx->send_root) { - ret = get_inode_info(sctx->send_root, clone_root->ino, NULL, - &gen, NULL, NULL, NULL, NULL); + ret = get_inode_gen(sctx->send_root, clone_root->ino, &gen); if (ret < 0) goto out; ret = get_cur_path(sctx, clone_root->ino, gen, p); @@ -4955,7 +5349,7 @@ static int send_clone(struct send_ctx *sctx, TLV_PUT_UUID(sctx, BTRFS_SEND_A_CLONE_UUID, clone_root->root->root_item.uuid); TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_CTRANSID, - le64_to_cpu(clone_root->root->root_item.ctransid)); + btrfs_root_ctransid(&clone_root->root->root_item)); TLV_PUT_PATH(sctx, BTRFS_SEND_A_CLONE_PATH, p); TLV_PUT_U64(sctx, BTRFS_SEND_A_CLONE_OFFSET, clone_root->offset); @@ -4977,38 +5371,64 @@ static int send_update_extent(struct send_ctx *sctx, int ret = 0; struct fs_path *p; - p = fs_path_alloc(); - if (!p) - return -ENOMEM; + p = get_cur_inode_path(sctx); + if (IS_ERR(p)) + return PTR_ERR(p); ret = begin_cmd(sctx, BTRFS_SEND_C_UPDATE_EXTENT); if (ret < 0) - goto out; + return ret; + + TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); + TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); + TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); - ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); + ret = send_cmd(sctx); + +tlv_put_failure: + return ret; +} + +static int send_fallocate(struct send_ctx *sctx, u32 mode, u64 offset, u64 len) +{ + struct fs_path *path; + int ret; + + path = get_cur_inode_path(sctx); + if (IS_ERR(path)) + return PTR_ERR(path); + + ret = begin_cmd(sctx, BTRFS_SEND_C_FALLOCATE); if (ret < 0) - goto out; + return ret; - TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); + TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, path); + TLV_PUT_U32(sctx, BTRFS_SEND_A_FALLOCATE_MODE, mode); TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); TLV_PUT_U64(sctx, BTRFS_SEND_A_SIZE, len); ret = send_cmd(sctx); tlv_put_failure: -out: - fs_path_free(p); return ret; } static int send_hole(struct send_ctx *sctx, u64 end) { struct fs_path *p = NULL; + u64 read_size = max_send_read_size(sctx); u64 offset = sctx->cur_inode_last_extent; - u64 len; int ret = 0; /* + * Starting with send stream v2 we have fallocate and can use it to + * punch holes instead of sending writes full of zeroes. + */ + if (proto_cmd_ok(sctx, BTRFS_SEND_C_FALLOCATE)) + return send_fallocate(sctx, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, + offset, end - offset); + + /* * A hole that starts at EOF or beyond it. Since we do not yet support * fallocate (for extent preallocation and hole punching), sending a * write of zeroes starting at EOF or beyond would later require issuing @@ -5017,25 +5437,32 @@ static int send_hole(struct send_ctx *sctx, u64 end) if (offset >= sctx->cur_inode_size) return 0; + /* + * Don't go beyond the inode's i_size due to prealloc extents that start + * after the i_size. + */ + end = min_t(u64, end, sctx->cur_inode_size); + if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) return send_update_extent(sctx, offset, end - offset); - p = fs_path_alloc(); - if (!p) - return -ENOMEM; - ret = get_cur_path(sctx, sctx->cur_ino, sctx->cur_inode_gen, p); - if (ret < 0) - goto tlv_put_failure; - memset(sctx->read_buf, 0, BTRFS_SEND_READ_SIZE); + p = get_cur_inode_path(sctx); + if (IS_ERR(p)) + return PTR_ERR(p); + while (offset < end) { - len = min_t(u64, end - offset, BTRFS_SEND_READ_SIZE); + u64 len = min(end - offset, read_size); ret = begin_cmd(sctx, BTRFS_SEND_C_WRITE); if (ret < 0) break; TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, p); TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); - TLV_PUT(sctx, BTRFS_SEND_A_DATA, sctx->read_buf, len); + ret = put_data_header(sctx, len); + if (ret < 0) + break; + memset(sctx->send_buf + sctx->send_size, 0, len); + sctx->send_size += len; ret = send_cmd(sctx); if (ret < 0) break; @@ -5043,45 +5470,330 @@ static int send_hole(struct send_ctx *sctx, u64 end) } sctx->cur_inode_next_write_offset = offset; tlv_put_failure: - fs_path_free(p); return ret; } -static int send_extent_data(struct send_ctx *sctx, - const u64 offset, - const u64 len) +static int send_encoded_inline_extent(struct send_ctx *sctx, + struct btrfs_path *path, u64 offset, + u64 len) +{ + struct btrfs_fs_info *fs_info = sctx->send_root->fs_info; + struct fs_path *fspath; + struct extent_buffer *leaf = path->nodes[0]; + struct btrfs_key key; + struct btrfs_file_extent_item *ei; + u64 ram_bytes; + size_t inline_size; + int ret; + + fspath = get_cur_inode_path(sctx); + if (IS_ERR(fspath)) + return PTR_ERR(fspath); + + ret = begin_cmd(sctx, BTRFS_SEND_C_ENCODED_WRITE); + if (ret < 0) + return ret; + + btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); + ram_bytes = btrfs_file_extent_ram_bytes(leaf, ei); + inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]); + + TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, fspath); + TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); + TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_FILE_LEN, + min(key.offset + ram_bytes - offset, len)); + TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_LEN, ram_bytes); + TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_OFFSET, offset - key.offset); + ret = btrfs_encoded_io_compression_from_extent(fs_info, + btrfs_file_extent_compression(leaf, ei)); + if (ret < 0) + return ret; + TLV_PUT_U32(sctx, BTRFS_SEND_A_COMPRESSION, ret); + + ret = put_data_header(sctx, inline_size); + if (ret < 0) + return ret; + read_extent_buffer(leaf, sctx->send_buf + sctx->send_size, + btrfs_file_extent_inline_start(ei), inline_size); + sctx->send_size += inline_size; + + ret = send_cmd(sctx); + +tlv_put_failure: + return ret; +} + +static int send_encoded_extent(struct send_ctx *sctx, struct btrfs_path *path, + u64 offset, u64 len) +{ + struct btrfs_root *root = sctx->send_root; + struct btrfs_fs_info *fs_info = root->fs_info; + struct btrfs_inode *inode; + struct fs_path *fspath; + struct extent_buffer *leaf = path->nodes[0]; + struct btrfs_key key; + struct btrfs_file_extent_item *ei; + u64 disk_bytenr, disk_num_bytes; + u32 data_offset; + struct btrfs_cmd_header *hdr; + u32 crc; + int ret; + + inode = btrfs_iget(sctx->cur_ino, root); + if (IS_ERR(inode)) + return PTR_ERR(inode); + + fspath = get_cur_inode_path(sctx); + if (IS_ERR(fspath)) { + ret = PTR_ERR(fspath); + goto out; + } + + ret = begin_cmd(sctx, BTRFS_SEND_C_ENCODED_WRITE); + if (ret < 0) + goto out; + + btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); + disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei); + disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, ei); + + TLV_PUT_PATH(sctx, BTRFS_SEND_A_PATH, fspath); + TLV_PUT_U64(sctx, BTRFS_SEND_A_FILE_OFFSET, offset); + TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_FILE_LEN, + min(key.offset + btrfs_file_extent_num_bytes(leaf, ei) - offset, + len)); + TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_LEN, + btrfs_file_extent_ram_bytes(leaf, ei)); + TLV_PUT_U64(sctx, BTRFS_SEND_A_UNENCODED_OFFSET, + offset - key.offset + btrfs_file_extent_offset(leaf, ei)); + ret = btrfs_encoded_io_compression_from_extent(fs_info, + btrfs_file_extent_compression(leaf, ei)); + if (ret < 0) + goto out; + TLV_PUT_U32(sctx, BTRFS_SEND_A_COMPRESSION, ret); + TLV_PUT_U32(sctx, BTRFS_SEND_A_ENCRYPTION, 0); + + ret = put_data_header(sctx, disk_num_bytes); + if (ret < 0) + goto out; + + /* + * We want to do I/O directly into the send buffer, so get the next page + * boundary in the send buffer. This means that there may be a gap + * between the beginning of the command and the file data. + */ + data_offset = PAGE_ALIGN(sctx->send_size); + if (unlikely(data_offset > sctx->send_max_size || + sctx->send_max_size - data_offset < disk_num_bytes)) { + ret = -EOVERFLOW; + goto out; + } + + /* + * Note that send_buf is a mapping of send_buf_pages, so this is really + * reading into send_buf. + */ + ret = btrfs_encoded_read_regular_fill_pages(inode, + disk_bytenr, disk_num_bytes, + sctx->send_buf_pages + + (data_offset >> PAGE_SHIFT), + NULL); + if (ret) + goto out; + + hdr = (struct btrfs_cmd_header *)sctx->send_buf; + hdr->len = cpu_to_le32(sctx->send_size + disk_num_bytes - sizeof(*hdr)); + hdr->crc = 0; + crc = crc32c(0, sctx->send_buf, sctx->send_size); + crc = crc32c(crc, sctx->send_buf + data_offset, disk_num_bytes); + hdr->crc = cpu_to_le32(crc); + + ret = write_buf(sctx->send_filp, sctx->send_buf, sctx->send_size, + &sctx->send_off); + if (!ret) { + ret = write_buf(sctx->send_filp, sctx->send_buf + data_offset, + disk_num_bytes, &sctx->send_off); + } + sctx->send_size = 0; + sctx->put_data = false; + +tlv_put_failure: +out: + iput(&inode->vfs_inode); + return ret; +} + +static int send_extent_data(struct send_ctx *sctx, struct btrfs_path *path, + const u64 offset, const u64 len) { + const u64 end = offset + len; + struct extent_buffer *leaf = path->nodes[0]; + struct btrfs_file_extent_item *ei; + u64 read_size = max_send_read_size(sctx); u64 sent = 0; if (sctx->flags & BTRFS_SEND_FLAG_NO_FILE_DATA) return send_update_extent(sctx, offset, len); + ei = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_file_extent_item); + if ((sctx->flags & BTRFS_SEND_FLAG_COMPRESSED) && + btrfs_file_extent_compression(leaf, ei) != BTRFS_COMPRESS_NONE) { + bool is_inline = (btrfs_file_extent_type(leaf, ei) == + BTRFS_FILE_EXTENT_INLINE); + + /* + * Send the compressed extent unless the compressed data is + * larger than the decompressed data. This can happen if we're + * not sending the entire extent, either because it has been + * partially overwritten/truncated or because this is a part of + * the extent that we couldn't clone in clone_range(). + */ + if (is_inline && + btrfs_file_extent_inline_item_len(leaf, + path->slots[0]) <= len) { + return send_encoded_inline_extent(sctx, path, offset, + len); + } else if (!is_inline && + btrfs_file_extent_disk_num_bytes(leaf, ei) <= len) { + return send_encoded_extent(sctx, path, offset, len); + } + } + + if (sctx->cur_inode == NULL) { + struct btrfs_inode *btrfs_inode; + struct btrfs_root *root = sctx->send_root; + + btrfs_inode = btrfs_iget(sctx->cur_ino, root); + if (IS_ERR(btrfs_inode)) + return PTR_ERR(btrfs_inode); + + sctx->cur_inode = &btrfs_inode->vfs_inode; + memset(&sctx->ra, 0, sizeof(struct file_ra_state)); + file_ra_state_init(&sctx->ra, sctx->cur_inode->i_mapping); + + /* + * It's very likely there are no pages from this inode in the page + * cache, so after reading extents and sending their data, we clean + * the page cache to avoid trashing the page cache (adding pressure + * to the page cache and forcing eviction of other data more useful + * for applications). + * + * We decide if we should clean the page cache simply by checking + * if the inode's mapping nrpages is 0 when we first open it, and + * not by using something like filemap_range_has_page() before + * reading an extent because when we ask the readahead code to + * read a given file range, it may (and almost always does) read + * pages from beyond that range (see the documentation for + * page_cache_sync_readahead()), so it would not be reliable, + * because after reading the first extent future calls to + * filemap_range_has_page() would return true because the readahead + * on the previous extent resulted in reading pages of the current + * extent as well. + */ + sctx->clean_page_cache = (sctx->cur_inode->i_mapping->nrpages == 0); + sctx->page_cache_clear_start = round_down(offset, PAGE_SIZE); + } + while (sent < len) { - u64 size = len - sent; + u64 size = min(len - sent, read_size); int ret; - if (size > BTRFS_SEND_READ_SIZE) - size = BTRFS_SEND_READ_SIZE; ret = send_write(sctx, offset + sent, size); if (ret < 0) return ret; - if (!ret) - break; - sent += ret; + sent += size; } + + if (sctx->clean_page_cache && PAGE_ALIGNED(end)) { + /* + * Always operate only on ranges that are a multiple of the page + * size. This is not only to prevent zeroing parts of a page in + * the case of subpage sector size, but also to guarantee we evict + * pages, as passing a range that is smaller than page size does + * not evict the respective page (only zeroes part of its content). + * + * Always start from the end offset of the last range cleared. + * This is because the readahead code may (and very often does) + * reads pages beyond the range we request for readahead. So if + * we have an extent layout like this: + * + * [ extent A ] [ extent B ] [ extent C ] + * + * When we ask page_cache_sync_readahead() to read extent A, it + * may also trigger reads for pages of extent B. If we are doing + * an incremental send and extent B has not changed between the + * parent and send snapshots, some or all of its pages may end + * up being read and placed in the page cache. So when truncating + * the page cache we always start from the end offset of the + * previously processed extent up to the end of the current + * extent. + */ + truncate_inode_pages_range(&sctx->cur_inode->i_data, + sctx->page_cache_clear_start, + end - 1); + sctx->page_cache_clear_start = end; + } + return 0; } -static int clone_range(struct send_ctx *sctx, - struct clone_root *clone_root, - const u64 disk_byte, - u64 data_offset, - u64 offset, - u64 len) +/* + * Search for a capability xattr related to sctx->cur_ino. If the capability is + * found, call send_set_xattr function to emit it. + * + * Return 0 if there isn't a capability, or when the capability was emitted + * successfully, or < 0 if an error occurred. + */ +static int send_capabilities(struct send_ctx *sctx) { - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); + struct btrfs_dir_item *di; + struct extent_buffer *leaf; + unsigned long data_ptr; + char AUTO_KFREE(buf); + int buf_len; + int ret = 0; + + path = alloc_path_for_send(); + if (!path) + return -ENOMEM; + + di = btrfs_lookup_xattr(NULL, sctx->send_root, path, sctx->cur_ino, + XATTR_NAME_CAPS, strlen(XATTR_NAME_CAPS), 0); + if (!di) { + /* There is no xattr for this inode */ + return 0; + } else if (IS_ERR(di)) { + return PTR_ERR(di); + } + + leaf = path->nodes[0]; + buf_len = btrfs_dir_data_len(leaf, di); + + buf = kmalloc(buf_len, GFP_KERNEL); + if (!buf) + return -ENOMEM; + + data_ptr = (unsigned long)(di + 1) + btrfs_dir_name_len(leaf, di); + read_extent_buffer(leaf, buf, data_ptr, buf_len); + + ret = send_set_xattr(sctx, XATTR_NAME_CAPS, + strlen(XATTR_NAME_CAPS), buf, buf_len); + return ret; +} + +static int clone_range(struct send_ctx *sctx, struct btrfs_path *dst_path, + struct clone_root *clone_root, const u64 disk_byte, + u64 data_offset, u64 offset, u64 len) +{ + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; int ret; + struct btrfs_inode_info info; + u64 clone_src_i_size = 0; /* * Prevent cloning from a zero offset with a length matching the sector @@ -5100,13 +5812,23 @@ static int clone_range(struct send_ctx *sctx, */ if (clone_root->offset == 0 && len == sctx->send_root->fs_info->sectorsize) - return send_extent_data(sctx, offset, len); + return send_extent_data(sctx, dst_path, offset, len); path = alloc_path_for_send(); if (!path) return -ENOMEM; /* + * There are inodes that have extents that lie behind its i_size. Don't + * accept clones from these extents. + */ + ret = get_inode_info(clone_root->root, clone_root->ino, &info); + btrfs_release_path(path); + if (ret < 0) + return ret; + clone_src_i_size = info.size; + + /* * We can't send a clone operation for the entire range if we find * extent items in the respective range in the source file that * refer to different extents or if we find holes. @@ -5133,7 +5855,7 @@ static int clone_range(struct send_ctx *sctx, key.offset = clone_root->offset; ret = btrfs_search_slot(NULL, clone_root->root, &key, path, 0, 0); if (ret < 0) - goto out; + return ret; if (ret > 0 && path->slots[0] > 0) { btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); if (key.objectid == clone_root->ino && @@ -5148,11 +5870,13 @@ static int clone_range(struct send_ctx *sctx, u8 type; u64 ext_len; u64 clone_len; + u64 clone_data_offset; + bool crossed_src_i_size = false; if (slot >= btrfs_header_nritems(leaf)) { ret = btrfs_next_leaf(clone_root->root, path); if (ret < 0) - goto out; + return ret; else if (ret > 0) break; continue; @@ -5186,9 +5910,10 @@ static int clone_range(struct send_ctx *sctx, if (hole_len > len) hole_len = len; - ret = send_extent_data(sctx, offset, hole_len); + ret = send_extent_data(sctx, dst_path, offset, + hole_len); if (ret < 0) - goto out; + return ret; len -= hole_len; if (len == 0) @@ -5201,33 +5926,129 @@ static int clone_range(struct send_ctx *sctx, if (key.offset >= clone_root->offset + len) break; + if (key.offset >= clone_src_i_size) + break; + + if (key.offset + ext_len > clone_src_i_size) { + ext_len = clone_src_i_size - key.offset; + crossed_src_i_size = true; + } + + clone_data_offset = btrfs_file_extent_offset(leaf, ei); + if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte) { + clone_root->offset = key.offset; + if (clone_data_offset < data_offset && + clone_data_offset + ext_len > data_offset) { + u64 extent_offset; + + extent_offset = data_offset - clone_data_offset; + ext_len -= extent_offset; + clone_data_offset += extent_offset; + clone_root->offset += extent_offset; + } + } + clone_len = min_t(u64, ext_len, len); if (btrfs_file_extent_disk_bytenr(leaf, ei) == disk_byte && - btrfs_file_extent_offset(leaf, ei) == data_offset) - ret = send_clone(sctx, offset, clone_len, clone_root); - else - ret = send_extent_data(sctx, offset, clone_len); + clone_data_offset == data_offset) { + const u64 src_end = clone_root->offset + clone_len; + const u64 sectorsize = SZ_64K; + + /* + * We can't clone the last block, when its size is not + * sector size aligned, into the middle of a file. If we + * do so, the receiver will get a failure (-EINVAL) when + * trying to clone or will silently corrupt the data in + * the destination file if it's on a kernel without the + * fix introduced by commit ac765f83f1397646 + * ("Btrfs: fix data corruption due to cloning of eof + * block). + * + * So issue a clone of the aligned down range plus a + * regular write for the eof block, if we hit that case. + * + * Also, we use the maximum possible sector size, 64K, + * because we don't know what's the sector size of the + * filesystem that receives the stream, so we have to + * assume the largest possible sector size. + */ + if (src_end == clone_src_i_size && + !IS_ALIGNED(src_end, sectorsize) && + offset + clone_len < sctx->cur_inode_size) { + u64 slen; + + slen = ALIGN_DOWN(src_end - clone_root->offset, + sectorsize); + if (slen > 0) { + ret = send_clone(sctx, offset, slen, + clone_root); + if (ret < 0) + return ret; + } + ret = send_extent_data(sctx, dst_path, + offset + slen, + clone_len - slen); + } else { + ret = send_clone(sctx, offset, clone_len, + clone_root); + } + } else if (crossed_src_i_size && clone_len < len) { + /* + * If we are at i_size of the clone source inode and we + * can not clone from it, terminate the loop. This is + * to avoid sending two write operations, one with a + * length matching clone_len and the final one after + * this loop with a length of len - clone_len. + * + * When using encoded writes (BTRFS_SEND_FLAG_COMPRESSED + * was passed to the send ioctl), this helps avoid + * sending an encoded write for an offset that is not + * sector size aligned, in case the i_size of the source + * inode is not sector size aligned. That will make the + * receiver fallback to decompression of the data and + * writing it using regular buffered IO, therefore while + * not incorrect, it's not optimal due decompression and + * possible re-compression at the receiver. + */ + break; + } else { + ret = send_extent_data(sctx, dst_path, offset, + clone_len); + } if (ret < 0) - goto out; + return ret; len -= clone_len; if (len == 0) break; offset += clone_len; clone_root->offset += clone_len; + + /* + * If we are cloning from the file we are currently processing, + * and using the send root as the clone root, we must stop once + * the current clone offset reaches the current eof of the file + * at the receiver, otherwise we would issue an invalid clone + * operation (source range going beyond eof) and cause the + * receiver to fail. So if we reach the current eof, bail out + * and fallback to a regular write. + */ + if (clone_root->root == sctx->send_root && + clone_root->ino == sctx->cur_ino && + clone_root->offset >= sctx->cur_inode_next_write_offset) + break; + data_offset += clone_len; next: path->slots[0]++; } if (len > 0) - ret = send_extent_data(sctx, offset, len); + ret = send_extent_data(sctx, dst_path, offset, len); else ret = 0; -out: - btrfs_free_path(path); return ret; } @@ -5237,51 +6058,76 @@ static int send_write_or_clone(struct send_ctx *sctx, struct clone_root *clone_root) { int ret = 0; - struct btrfs_file_extent_item *ei; u64 offset = key->offset; - u64 len; - u8 type; - u64 bs = sctx->send_root->fs_info->sb->s_blocksize; + u64 end; + u64 bs = sctx->send_root->fs_info->sectorsize; + struct btrfs_file_extent_item *ei; + u64 disk_byte; + u64 data_offset; + u64 num_bytes; + struct btrfs_inode_info info = { 0 }; - ei = btrfs_item_ptr(path->nodes[0], path->slots[0], - struct btrfs_file_extent_item); - type = btrfs_file_extent_type(path->nodes[0], ei); - if (type == BTRFS_FILE_EXTENT_INLINE) { - len = btrfs_file_extent_ram_bytes(path->nodes[0], ei); + end = min_t(u64, btrfs_file_extent_end(path), sctx->cur_inode_size); + if (offset >= end) + return 0; + + num_bytes = end - offset; + + if (!clone_root) + goto write_data; + + if (IS_ALIGNED(end, bs)) + goto clone_data; + + /* + * If the extent end is not aligned, we can clone if the extent ends at + * the i_size of the inode and the clone range ends at the i_size of the + * source inode, otherwise the clone operation fails with -EINVAL. + */ + if (end != sctx->cur_inode_size) + goto write_data; + + ret = get_inode_info(clone_root->root, clone_root->ino, &info); + if (ret < 0) + return ret; + + if (clone_root->offset + num_bytes == info.size) { /* - * it is possible the inline item won't cover the whole page, - * but there may be items after this page. Make - * sure to send the whole thing + * The final size of our file matches the end offset, but it may + * be that its current size is larger, so we have to truncate it + * to any value between the start offset of the range and the + * final i_size, otherwise the clone operation is invalid + * because it's unaligned and it ends before the current EOF. + * We do this truncate to the final i_size when we finish + * processing the inode, but it's too late by then. And here we + * truncate to the start offset of the range because it's always + * sector size aligned while if it were the final i_size it + * would result in dirtying part of a page, filling part of a + * page with zeroes and then having the clone operation at the + * receiver trigger IO and wait for it due to the dirty page. */ - len = PAGE_ALIGN(len); - } else { - len = btrfs_file_extent_num_bytes(path->nodes[0], ei); - } - - if (offset >= sctx->cur_inode_size) { - ret = 0; - goto out; - } - if (offset + len > sctx->cur_inode_size) - len = sctx->cur_inode_size - offset; - if (len == 0) { - ret = 0; - goto out; + if (sctx->parent_root != NULL) { + ret = send_truncate(sctx, sctx->cur_ino, + sctx->cur_inode_gen, offset); + if (ret < 0) + return ret; + } + goto clone_data; } - if (clone_root && IS_ALIGNED(offset + len, bs)) { - u64 disk_byte; - u64 data_offset; +write_data: + ret = send_extent_data(sctx, path, offset, num_bytes); + sctx->cur_inode_next_write_offset = end; + return ret; - disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei); - data_offset = btrfs_file_extent_offset(path->nodes[0], ei); - ret = clone_range(sctx, clone_root, disk_byte, data_offset, - offset, len); - } else { - ret = send_extent_data(sctx, offset, len); - } - sctx->cur_inode_next_write_offset = offset + len; -out: +clone_data: + ei = btrfs_item_ptr(path->nodes[0], path->slots[0], + struct btrfs_file_extent_item); + disk_byte = btrfs_file_extent_disk_bytenr(path->nodes[0], ei); + data_offset = btrfs_file_extent_offset(path->nodes[0], ei); + ret = clone_range(sctx, path, clone_root, disk_byte, data_offset, offset, + num_bytes); + sctx->cur_inode_next_write_offset = end; return ret; } @@ -5291,7 +6137,7 @@ static int is_extent_unchanged(struct send_ctx *sctx, { int ret = 0; struct btrfs_key key; - struct btrfs_path *path = NULL; + BTRFS_PATH_AUTO_FREE(path); struct extent_buffer *eb; int slot; struct btrfs_key found_key; @@ -5317,10 +6163,9 @@ static int is_extent_unchanged(struct send_ctx *sctx, ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); left_type = btrfs_file_extent_type(eb, ei); - if (left_type != BTRFS_FILE_EXTENT_REG) { - ret = 0; - goto out; - } + if (left_type != BTRFS_FILE_EXTENT_REG) + return 0; + left_disknr = btrfs_file_extent_disk_bytenr(eb, ei); left_len = btrfs_file_extent_num_bytes(eb, ei); left_offset = btrfs_file_extent_offset(eb, ei); @@ -5352,11 +6197,9 @@ static int is_extent_unchanged(struct send_ctx *sctx, key.offset = ekey->offset; ret = btrfs_search_slot_for_read(sctx->parent_root, &key, path, 0, 0); if (ret < 0) - goto out; - if (ret) { - ret = 0; - goto out; - } + return ret; + if (ret) + return 0; /* * Handle special case where the right side has no extents at all. @@ -5365,11 +6208,9 @@ static int is_extent_unchanged(struct send_ctx *sctx, slot = path->slots[0]; btrfs_item_key_to_cpu(eb, &found_key, slot); if (found_key.objectid != key.objectid || - found_key.type != key.type) { + found_key.type != key.type) /* If we're a hole then just pretend nothing changed */ - ret = (left_disknr) ? 0 : 1; - goto out; - } + return (left_disknr ? 0 : 1); /* * We're now on 2a, 2b or 7. @@ -5379,10 +6220,8 @@ static int is_extent_unchanged(struct send_ctx *sctx, ei = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); right_type = btrfs_file_extent_type(eb, ei); if (right_type != BTRFS_FILE_EXTENT_REG && - right_type != BTRFS_FILE_EXTENT_INLINE) { - ret = 0; - goto out; - } + right_type != BTRFS_FILE_EXTENT_INLINE) + return 0; if (right_type == BTRFS_FILE_EXTENT_INLINE) { right_len = btrfs_file_extent_ram_bytes(eb, ei); @@ -5395,11 +6234,9 @@ static int is_extent_unchanged(struct send_ctx *sctx, * Are we at extent 8? If yes, we know the extent is changed. * This may only happen on the first iteration. */ - if (found_key.offset + right_len <= ekey->offset) { + if (found_key.offset + right_len <= ekey->offset) /* If we're a hole just pretend nothing changed */ - ret = (left_disknr) ? 0 : 1; - goto out; - } + return (left_disknr ? 0 : 1); /* * We just wanted to see if when we have an inline extent, what @@ -5409,10 +6246,8 @@ static int is_extent_unchanged(struct send_ctx *sctx, * compressed extent representing data with a size matching * the page size (currently the same as sector size). */ - if (right_type == BTRFS_FILE_EXTENT_INLINE) { - ret = 0; - goto out; - } + if (right_type == BTRFS_FILE_EXTENT_INLINE) + return 0; right_disknr = btrfs_file_extent_disk_bytenr(eb, ei); right_offset = btrfs_file_extent_offset(eb, ei); @@ -5432,17 +6267,15 @@ static int is_extent_unchanged(struct send_ctx *sctx, */ if (left_disknr != right_disknr || left_offset_fixed != right_offset || - left_gen != right_gen) { - ret = 0; - goto out; - } + left_gen != right_gen) + return 0; /* * Go to the next extent. */ ret = btrfs_next_item(sctx->parent_root, path); if (ret < 0) - goto out; + return ret; if (!ret) { eb = path->nodes[0]; slot = path->slots[0]; @@ -5453,10 +6286,9 @@ static int is_extent_unchanged(struct send_ctx *sctx, key.offset += right_len; break; } - if (found_key.offset != key.offset + right_len) { - ret = 0; - goto out; - } + if (found_key.offset != key.offset + right_len) + return 0; + key = found_key; } @@ -5469,20 +6301,14 @@ static int is_extent_unchanged(struct send_ctx *sctx, else ret = 0; - -out: - btrfs_free_path(path); return ret; } static int get_last_extent(struct send_ctx *sctx, u64 offset) { - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_root *root = sctx->send_root; - struct btrfs_file_extent_item *fi; struct btrfs_key key; - u64 extent_end; - u8 type; int ret; path = alloc_path_for_send(); @@ -5496,26 +6322,13 @@ static int get_last_extent(struct send_ctx *sctx, u64 offset) key.offset = offset; ret = btrfs_search_slot_for_read(root, &key, path, 0, 1); if (ret < 0) - goto out; + return ret; ret = 0; btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); if (key.objectid != sctx->cur_ino || key.type != BTRFS_EXTENT_DATA_KEY) - goto out; + return ret; - fi = btrfs_item_ptr(path->nodes[0], path->slots[0], - struct btrfs_file_extent_item); - type = btrfs_file_extent_type(path->nodes[0], fi); - if (type == BTRFS_FILE_EXTENT_INLINE) { - u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi); - extent_end = ALIGN(key.offset + size, - sctx->send_root->fs_info->sectorsize); - } else { - extent_end = key.offset + - btrfs_file_extent_num_bytes(path->nodes[0], fi); - } - sctx->cur_inode_last_extent = extent_end; -out: - btrfs_free_path(path); + sctx->cur_inode_last_extent = btrfs_file_extent_end(path); return ret; } @@ -5523,7 +6336,7 @@ static int range_is_hole_in_parent(struct send_ctx *sctx, const u64 start, const u64 end) { - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; struct btrfs_root *root = sctx->parent_root; u64 search_start = start; @@ -5538,7 +6351,7 @@ static int range_is_hole_in_parent(struct send_ctx *sctx, key.offset = search_start; ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); if (ret < 0) - goto out; + return ret; if (ret > 0 && path->slots[0] > 0) path->slots[0]--; @@ -5551,8 +6364,8 @@ static int range_is_hole_in_parent(struct send_ctx *sctx, if (slot >= btrfs_header_nritems(leaf)) { ret = btrfs_next_leaf(root, path); if (ret < 0) - goto out; - else if (ret > 0) + return ret; + if (ret > 0) break; continue; } @@ -5567,71 +6380,40 @@ static int range_is_hole_in_parent(struct send_ctx *sctx, break; fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); - if (btrfs_file_extent_type(leaf, fi) == - BTRFS_FILE_EXTENT_INLINE) { - u64 size = btrfs_file_extent_ram_bytes(leaf, fi); - - extent_end = ALIGN(key.offset + size, - root->fs_info->sectorsize); - } else { - extent_end = key.offset + - btrfs_file_extent_num_bytes(leaf, fi); - } + extent_end = btrfs_file_extent_end(path); if (extent_end <= start) goto next; if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0) { search_start = extent_end; goto next; } - ret = 0; - goto out; + return 0; next: path->slots[0]++; } - ret = 1; -out: - btrfs_free_path(path); - return ret; + return 1; } static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, struct btrfs_key *key) { - struct btrfs_file_extent_item *fi; - u64 extent_end; - u8 type; int ret = 0; if (sctx->cur_ino != key->objectid || !need_send_hole(sctx)) return 0; - if (sctx->cur_inode_last_extent == (u64)-1) { - ret = get_last_extent(sctx, key->offset - 1); - if (ret) - return ret; - } - - fi = btrfs_item_ptr(path->nodes[0], path->slots[0], - struct btrfs_file_extent_item); - type = btrfs_file_extent_type(path->nodes[0], fi); - if (type == BTRFS_FILE_EXTENT_INLINE) { - u64 size = btrfs_file_extent_ram_bytes(path->nodes[0], fi); - extent_end = ALIGN(key->offset + size, - sctx->send_root->fs_info->sectorsize); - } else { - extent_end = key->offset + - btrfs_file_extent_num_bytes(path->nodes[0], fi); - } - - if (path->slots[0] == 0 && - sctx->cur_inode_last_extent < key->offset) { - /* - * We might have skipped entire leafs that contained only - * file extent items for our current inode. These leafs have - * a generation number smaller (older) than the one in the - * current leaf and the leaf our last extent came from, and - * are located between these 2 leafs. - */ + /* + * Get last extent's end offset (exclusive) if we haven't determined it + * yet (we're processing the first file extent item that is new), or if + * we're at the first slot of a leaf and the last extent's end is less + * than the current extent's offset, because we might have skipped + * entire leaves that contained only file extent items for our current + * inode. These leaves have a generation number smaller (older) than the + * one in the current leaf and the leaf our last extent came from, and + * are located between these 2 leaves. + */ + if ((sctx->cur_inode_last_extent == (u64)-1) || + (path->slots[0] == 0 && sctx->cur_inode_last_extent < key->offset)) { ret = get_last_extent(sctx, key->offset - 1); if (ret) return ret; @@ -5648,7 +6430,7 @@ static int maybe_send_hole(struct send_ctx *sctx, struct btrfs_path *path, else ret = 0; } - sctx->cur_inode_last_extent = extent_end; + sctx->cur_inode_last_extent = btrfs_file_extent_end(path); return ret; } @@ -5714,13 +6496,12 @@ out: static int process_all_extents(struct send_ctx *sctx) { - int ret; + int ret = 0; + int iter_ret = 0; struct btrfs_root *root; - struct btrfs_path *path; + BTRFS_PATH_AUTO_FREE(path); struct btrfs_key key; struct btrfs_key found_key; - struct extent_buffer *eb; - int slot; root = sctx->send_root; path = alloc_path_for_send(); @@ -5730,46 +6511,25 @@ static int process_all_extents(struct send_ctx *sctx) key.objectid = sctx->cmp_key->objectid; key.type = BTRFS_EXTENT_DATA_KEY; key.offset = 0; - ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); - if (ret < 0) - goto out; - - while (1) { - eb = path->nodes[0]; - slot = path->slots[0]; - - if (slot >= btrfs_header_nritems(eb)) { - ret = btrfs_next_leaf(root, path); - if (ret < 0) { - goto out; - } else if (ret > 0) { - ret = 0; - break; - } - continue; - } - - btrfs_item_key_to_cpu(eb, &found_key, slot); - + btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { if (found_key.objectid != key.objectid || found_key.type != key.type) { ret = 0; - goto out; + break; } ret = process_extent(sctx, path, &found_key); if (ret < 0) - goto out; - - path->slots[0]++; + break; } + /* Catch error found during iteration */ + if (iter_ret < 0) + ret = iter_ret; -out: - btrfs_free_path(path); return ret; } -static int process_recorded_refs_if_needed(struct send_ctx *sctx, int at_end, +static int process_recorded_refs_if_needed(struct send_ctx *sctx, bool at_end, int *pending_move, int *refs_processed) { @@ -5792,17 +6552,21 @@ out: return ret; } -static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) +static int finish_inode_if_needed(struct send_ctx *sctx, bool at_end) { int ret = 0; + struct btrfs_inode_info info; u64 left_mode; u64 left_uid; u64 left_gid; + u64 left_fileattr; u64 right_mode; u64 right_uid; u64 right_gid; + u64 right_fileattr; int need_chmod = 0; int need_chown = 0; + bool need_fileattr = false; int need_truncate = 1; int pending_move = 0; int refs_processed = 0; @@ -5834,11 +6598,13 @@ static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) goto out; if (!at_end && sctx->cmp_key->objectid == sctx->cur_ino) goto out; - - ret = get_inode_info(sctx->send_root, sctx->cur_ino, NULL, NULL, - &left_mode, &left_uid, &left_gid, NULL); + ret = get_inode_info(sctx->send_root, sctx->cur_ino, &info); if (ret < 0) goto out; + left_mode = info.mode; + left_uid = info.uid; + left_gid = info.gid; + left_fileattr = info.fileattr; if (!sctx->parent_root || sctx->cur_inode_new) { need_chown = 1; @@ -5849,16 +6615,21 @@ static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) } else { u64 old_size; - ret = get_inode_info(sctx->parent_root, sctx->cur_ino, - &old_size, NULL, &right_mode, &right_uid, - &right_gid, NULL); + ret = get_inode_info(sctx->parent_root, sctx->cur_ino, &info); if (ret < 0) goto out; + old_size = info.size; + right_mode = info.mode; + right_uid = info.uid; + right_gid = info.gid; + right_fileattr = info.fileattr; if (left_uid != right_uid || left_gid != right_gid) need_chown = 1; if (!S_ISLNK(sctx->cur_inode_mode) && left_mode != right_mode) need_chmod = 1; + if (!S_ISLNK(sctx->cur_inode_mode) && left_fileattr != right_fileattr) + need_fileattr = true; if ((old_size == sctx->cur_inode_size) || (sctx->cur_inode_size > old_size && sctx->cur_inode_next_write_offset == sctx->cur_inode_size)) @@ -5874,11 +6645,20 @@ static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) if (ret) goto out; } - if (sctx->cur_inode_last_extent < - sctx->cur_inode_size) { - ret = send_hole(sctx, sctx->cur_inode_size); - if (ret) + if (sctx->cur_inode_last_extent < sctx->cur_inode_size) { + ret = range_is_hole_in_parent(sctx, + sctx->cur_inode_last_extent, + sctx->cur_inode_size); + if (ret < 0) { goto out; + } else if (ret == 0) { + ret = send_hole(sctx, sctx->cur_inode_size); + if (ret < 0) + goto out; + } else { + /* Range is already a hole, skip. */ + ret = 0; + } } } if (need_truncate) { @@ -5902,6 +6682,23 @@ static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) if (ret < 0) goto out; } + if (need_fileattr) { + ret = send_fileattr(sctx, sctx->cur_ino, sctx->cur_inode_gen, + left_fileattr); + if (ret < 0) + goto out; + } + + if (proto_cmd_ok(sctx, BTRFS_SEND_C_ENABLE_VERITY) + && sctx->cur_inode_needs_verity) { + ret = process_verity(sctx); + if (ret < 0) + goto out; + } + + ret = send_capabilities(sctx); + if (ret < 0) + goto out; /* * If other directory inodes depended on our current directory @@ -5919,100 +6716,51 @@ static int finish_inode_if_needed(struct send_ctx *sctx, int at_end) * it's moved/renamed, therefore we don't need to do it here. */ sctx->send_progress = sctx->cur_ino + 1; - ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); + + /* + * If the current inode is a non-empty directory, delay issuing + * the utimes command for it, as it's very likely we have inodes + * with an higher number inside it. We want to issue the utimes + * command only after adding all dentries to it. + */ + if (S_ISDIR(sctx->cur_inode_mode) && sctx->cur_inode_size > 0) + ret = cache_dir_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); + else + ret = send_utimes(sctx, sctx->cur_ino, sctx->cur_inode_gen); + if (ret < 0) goto out; } out: - return ret; -} - -struct parent_paths_ctx { - struct list_head *refs; - struct send_ctx *sctx; -}; - -static int record_parent_ref(int num, u64 dir, int index, struct fs_path *name, - void *ctx) -{ - struct parent_paths_ctx *ppctx = ctx; + if (!ret) + ret = trim_dir_utimes_cache(sctx); - return record_ref(ppctx->sctx->parent_root, dir, name, ppctx->sctx, - ppctx->refs); + return ret; } -/* - * Issue unlink operations for all paths of the current inode found in the - * parent snapshot. - */ -static int btrfs_unlink_all_paths(struct send_ctx *sctx) +static void close_current_inode(struct send_ctx *sctx) { - LIST_HEAD(deleted_refs); - struct btrfs_path *path; - struct btrfs_key key; - struct parent_paths_ctx ctx; - int ret; - - path = alloc_path_for_send(); - if (!path) - return -ENOMEM; - - key.objectid = sctx->cur_ino; - key.type = BTRFS_INODE_REF_KEY; - key.offset = 0; - ret = btrfs_search_slot(NULL, sctx->parent_root, &key, path, 0, 0); - if (ret < 0) - goto out; - - ctx.refs = &deleted_refs; - ctx.sctx = sctx; - - while (true) { - struct extent_buffer *eb = path->nodes[0]; - int slot = path->slots[0]; - - if (slot >= btrfs_header_nritems(eb)) { - ret = btrfs_next_leaf(sctx->parent_root, path); - if (ret < 0) - goto out; - else if (ret > 0) - break; - continue; - } - - btrfs_item_key_to_cpu(eb, &key, slot); - if (key.objectid != sctx->cur_ino) - break; - if (key.type != BTRFS_INODE_REF_KEY && - key.type != BTRFS_INODE_EXTREF_KEY) - break; + u64 i_size; - ret = iterate_inode_ref(sctx->parent_root, path, &key, 1, - record_parent_ref, &ctx); - if (ret < 0) - goto out; + if (sctx->cur_inode == NULL) + return; - path->slots[0]++; - } + i_size = i_size_read(sctx->cur_inode); - while (!list_empty(&deleted_refs)) { - struct recorded_ref *ref; + /* + * If we are doing an incremental send, we may have extents between the + * last processed extent and the i_size that have not been processed + * because they haven't changed but we may have read some of their pages + * through readahead, see the comments at send_extent_data(). + */ + if (sctx->clean_page_cache && sctx->page_cache_clear_start < i_size) + truncate_inode_pages_range(&sctx->cur_inode->i_data, + sctx->page_cache_clear_start, + round_up(i_size, PAGE_SIZE) - 1); - ref = list_first_entry(&deleted_refs, struct recorded_ref, list); - ret = send_unlink(sctx, ref->full_path); - if (ret < 0) - goto out; - fs_path_free(ref->full_path); - list_del(&ref->list); - kfree(ref); - } - ret = 0; -out: - btrfs_free_path(path); - if (ret) - __free_recorded_refs(&deleted_refs); - return ret; + iput(sctx->cur_inode); + sctx->cur_inode = NULL; } static int changed_inode(struct send_ctx *sctx, @@ -6025,11 +6773,14 @@ static int changed_inode(struct send_ctx *sctx, u64 left_gen = 0; u64 right_gen = 0; + close_current_inode(sctx); + sctx->cur_ino = key->objectid; - sctx->cur_inode_new_gen = 0; + sctx->cur_inode_new_gen = false; sctx->cur_inode_last_extent = (u64)-1; sctx->cur_inode_next_write_offset = 0; sctx->ignore_cur_inode = false; + fs_path_reset(&sctx->cur_inode_path); /* * Set send_progress to current inode. This will tell all get_cur_xxx @@ -6067,7 +6818,7 @@ static int changed_inode(struct send_ctx *sctx, */ if (left_gen != right_gen && sctx->cur_ino != BTRFS_FIRST_FREE_OBJECTID) - sctx->cur_inode_new_gen = 1; + sctx->cur_inode_new_gen = true; } /* @@ -6079,28 +6830,39 @@ static int changed_inode(struct send_ctx *sctx, * file descriptor against it or turning a RO snapshot into RW mode, * keep an open file descriptor against a file, delete it and then * turn the snapshot back to RO mode before using it for a send - * operation. So if we find such cases, ignore the inode and all its - * items completely if it's a new inode, or if it's a changed inode - * make sure all its previous paths (from the parent snapshot) are all - * unlinked and all other the inode items are ignored. + * operation. The former is what the receiver operation does. + * Therefore, if we want to send these snapshots soon after they're + * received, we need to handle orphan inodes as well. Moreover, orphans + * can appear not only in the send snapshot but also in the parent + * snapshot. Here are several cases: + * + * Case 1: BTRFS_COMPARE_TREE_NEW + * | send snapshot | action + * -------------------------------- + * nlink | 0 | ignore + * + * Case 2: BTRFS_COMPARE_TREE_DELETED + * | parent snapshot | action + * ---------------------------------- + * nlink | 0 | as usual + * Note: No unlinks will be sent because there're no paths for it. + * + * Case 3: BTRFS_COMPARE_TREE_CHANGED + * | | parent snapshot | send snapshot | action + * ----------------------------------------------------------------------- + * subcase 1 | nlink | 0 | 0 | ignore + * subcase 2 | nlink | >0 | 0 | new_gen(deletion) + * subcase 3 | nlink | 0 | >0 | new_gen(creation) + * */ - if (result == BTRFS_COMPARE_TREE_NEW || - result == BTRFS_COMPARE_TREE_CHANGED) { - u32 nlinks; - - nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii); - if (nlinks == 0) { + if (result == BTRFS_COMPARE_TREE_NEW) { + if (btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii) == 0) { sctx->ignore_cur_inode = true; - if (result == BTRFS_COMPARE_TREE_CHANGED) - ret = btrfs_unlink_all_paths(sctx); goto out; } - } - - if (result == BTRFS_COMPARE_TREE_NEW) { sctx->cur_inode_gen = left_gen; - sctx->cur_inode_new = 1; - sctx->cur_inode_deleted = 0; + sctx->cur_inode_new = true; + sctx->cur_inode_deleted = false; sctx->cur_inode_size = btrfs_inode_size( sctx->left_path->nodes[0], left_ii); sctx->cur_inode_mode = btrfs_inode_mode( @@ -6111,13 +6873,23 @@ static int changed_inode(struct send_ctx *sctx, ret = send_create_inode_if_needed(sctx); } else if (result == BTRFS_COMPARE_TREE_DELETED) { sctx->cur_inode_gen = right_gen; - sctx->cur_inode_new = 0; - sctx->cur_inode_deleted = 1; + sctx->cur_inode_new = false; + sctx->cur_inode_deleted = true; sctx->cur_inode_size = btrfs_inode_size( sctx->right_path->nodes[0], right_ii); sctx->cur_inode_mode = btrfs_inode_mode( sctx->right_path->nodes[0], right_ii); } else if (result == BTRFS_COMPARE_TREE_CHANGED) { + u32 new_nlinks, old_nlinks; + + new_nlinks = btrfs_inode_nlink(sctx->left_path->nodes[0], left_ii); + old_nlinks = btrfs_inode_nlink(sctx->right_path->nodes[0], right_ii); + if (new_nlinks == 0 && old_nlinks == 0) { + sctx->ignore_cur_inode = true; + goto out; + } else if (new_nlinks == 0 || old_nlinks == 0) { + sctx->cur_inode_new_gen = 1; + } /* * We need to do some special handling in case the inode was * reported as changed with a changed generation number. This @@ -6129,58 +6901,66 @@ static int changed_inode(struct send_ctx *sctx, /* * First, process the inode as if it was deleted. */ - sctx->cur_inode_gen = right_gen; - sctx->cur_inode_new = 0; - sctx->cur_inode_deleted = 1; - sctx->cur_inode_size = btrfs_inode_size( - sctx->right_path->nodes[0], right_ii); - sctx->cur_inode_mode = btrfs_inode_mode( - sctx->right_path->nodes[0], right_ii); - ret = process_all_refs(sctx, - BTRFS_COMPARE_TREE_DELETED); - if (ret < 0) - goto out; + if (old_nlinks > 0) { + sctx->cur_inode_gen = right_gen; + sctx->cur_inode_new = false; + sctx->cur_inode_deleted = true; + sctx->cur_inode_size = btrfs_inode_size( + sctx->right_path->nodes[0], right_ii); + sctx->cur_inode_mode = btrfs_inode_mode( + sctx->right_path->nodes[0], right_ii); + ret = process_all_refs(sctx, + BTRFS_COMPARE_TREE_DELETED); + if (ret < 0) + goto out; + } /* * Now process the inode as if it was new. */ - sctx->cur_inode_gen = left_gen; - sctx->cur_inode_new = 1; - sctx->cur_inode_deleted = 0; - sctx->cur_inode_size = btrfs_inode_size( - sctx->left_path->nodes[0], left_ii); - sctx->cur_inode_mode = btrfs_inode_mode( - sctx->left_path->nodes[0], left_ii); - sctx->cur_inode_rdev = btrfs_inode_rdev( - sctx->left_path->nodes[0], left_ii); - ret = send_create_inode_if_needed(sctx); - if (ret < 0) - goto out; + if (new_nlinks > 0) { + sctx->cur_inode_gen = left_gen; + sctx->cur_inode_new = true; + sctx->cur_inode_deleted = false; + sctx->cur_inode_size = btrfs_inode_size( + sctx->left_path->nodes[0], + left_ii); + sctx->cur_inode_mode = btrfs_inode_mode( + sctx->left_path->nodes[0], + left_ii); + sctx->cur_inode_rdev = btrfs_inode_rdev( + sctx->left_path->nodes[0], + left_ii); + ret = send_create_inode_if_needed(sctx); + if (ret < 0) + goto out; - ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); - if (ret < 0) - goto out; - /* - * Advance send_progress now as we did not get into - * process_recorded_refs_if_needed in the new_gen case. - */ - sctx->send_progress = sctx->cur_ino + 1; + ret = process_all_refs(sctx, BTRFS_COMPARE_TREE_NEW); + if (ret < 0) + goto out; + /* + * Advance send_progress now as we did not get + * into process_recorded_refs_if_needed in the + * new_gen case. + */ + sctx->send_progress = sctx->cur_ino + 1; - /* - * Now process all extents and xattrs of the inode as if - * they were all new. - */ - ret = process_all_extents(sctx); - if (ret < 0) - goto out; - ret = process_all_new_xattrs(sctx); - if (ret < 0) - goto out; + /* + * Now process all extents and xattrs of the + * inode as if they were all new. + */ + ret = process_all_extents(sctx); + if (ret < 0) + goto out; + ret = process_all_new_xattrs(sctx); + if (ret < 0) + goto out; + } } else { sctx->cur_inode_gen = left_gen; - sctx->cur_inode_new = 0; - sctx->cur_inode_new_gen = 0; - sctx->cur_inode_deleted = 0; + sctx->cur_inode_new = false; + sctx->cur_inode_new_gen = false; + sctx->cur_inode_deleted = false; sctx->cur_inode_size = btrfs_inode_size( sctx->left_path->nodes[0], left_ii); sctx->cur_inode_mode = btrfs_inode_mode( @@ -6207,7 +6987,7 @@ static int changed_ref(struct send_ctx *sctx, { int ret = 0; - if (sctx->cur_ino != sctx->cmp_key->objectid) { + if (unlikely(sctx->cur_ino != sctx->cmp_key->objectid)) { inconsistent_snapshot_error(sctx, result, "reference"); return -EIO; } @@ -6235,7 +7015,7 @@ static int changed_xattr(struct send_ctx *sctx, { int ret = 0; - if (sctx->cur_ino != sctx->cmp_key->objectid) { + if (unlikely(sctx->cur_ino != sctx->cmp_key->objectid)) { inconsistent_snapshot_error(sctx, result, "xattr"); return -EIO; } @@ -6262,68 +7042,21 @@ static int changed_extent(struct send_ctx *sctx, { int ret = 0; - if (sctx->cur_ino != sctx->cmp_key->objectid) { - - if (result == BTRFS_COMPARE_TREE_CHANGED) { - struct extent_buffer *leaf_l; - struct extent_buffer *leaf_r; - struct btrfs_file_extent_item *ei_l; - struct btrfs_file_extent_item *ei_r; - - leaf_l = sctx->left_path->nodes[0]; - leaf_r = sctx->right_path->nodes[0]; - ei_l = btrfs_item_ptr(leaf_l, - sctx->left_path->slots[0], - struct btrfs_file_extent_item); - ei_r = btrfs_item_ptr(leaf_r, - sctx->right_path->slots[0], - struct btrfs_file_extent_item); - - /* - * We may have found an extent item that has changed - * only its disk_bytenr field and the corresponding - * inode item was not updated. This case happens due to - * very specific timings during relocation when a leaf - * that contains file extent items is COWed while - * relocation is ongoing and its in the stage where it - * updates data pointers. So when this happens we can - * safely ignore it since we know it's the same extent, - * but just at different logical and physical locations - * (when an extent is fully replaced with a new one, we - * know the generation number must have changed too, - * since snapshot creation implies committing the current - * transaction, and the inode item must have been updated - * as well). - * This replacement of the disk_bytenr happens at - * relocation.c:replace_file_extents() through - * relocation.c:btrfs_reloc_cow_block(). - */ - if (btrfs_file_extent_generation(leaf_l, ei_l) == - btrfs_file_extent_generation(leaf_r, ei_r) && - btrfs_file_extent_ram_bytes(leaf_l, ei_l) == - btrfs_file_extent_ram_bytes(leaf_r, ei_r) && - btrfs_file_extent_compression(leaf_l, ei_l) == - btrfs_file_extent_compression(leaf_r, ei_r) && - btrfs_file_extent_encryption(leaf_l, ei_l) == - btrfs_file_extent_encryption(leaf_r, ei_r) && - btrfs_file_extent_other_encoding(leaf_l, ei_l) == - btrfs_file_extent_other_encoding(leaf_r, ei_r) && - btrfs_file_extent_type(leaf_l, ei_l) == - btrfs_file_extent_type(leaf_r, ei_r) && - btrfs_file_extent_disk_bytenr(leaf_l, ei_l) != - btrfs_file_extent_disk_bytenr(leaf_r, ei_r) && - btrfs_file_extent_disk_num_bytes(leaf_l, ei_l) == - btrfs_file_extent_disk_num_bytes(leaf_r, ei_r) && - btrfs_file_extent_offset(leaf_l, ei_l) == - btrfs_file_extent_offset(leaf_r, ei_r) && - btrfs_file_extent_num_bytes(leaf_l, ei_l) == - btrfs_file_extent_num_bytes(leaf_r, ei_r)) - return 0; - } - - inconsistent_snapshot_error(sctx, result, "extent"); - return -EIO; - } + /* + * We have found an extent item that changed without the inode item + * having changed. This can happen either after relocation (where the + * disk_bytenr of an extent item is replaced at + * relocation.c:replace_file_extents()) or after deduplication into a + * file in both the parent and send snapshots (where an extent item can + * get modified or replaced with a new one). Note that deduplication + * updates the inode item, but it only changes the iversion (sequence + * field in the inode item) of the inode, so if a file is deduplicated + * the same amount of times in both the parent and send snapshots, its + * iversion becomes the same in both snapshots, whence the inode item is + * the same on both snapshots. + */ + if (sctx->cur_ino != sctx->cmp_key->objectid) + return 0; if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { if (result != BTRFS_COMPARE_TREE_DELETED) @@ -6334,18 +7067,25 @@ static int changed_extent(struct send_ctx *sctx, return ret; } +static int changed_verity(struct send_ctx *sctx, enum btrfs_compare_tree_result result) +{ + if (!sctx->cur_inode_new_gen && !sctx->cur_inode_deleted) { + if (result == BTRFS_COMPARE_TREE_NEW) + sctx->cur_inode_needs_verity = true; + } + return 0; +} + static int dir_changed(struct send_ctx *sctx, u64 dir) { u64 orig_gen, new_gen; int ret; - ret = get_inode_info(sctx->send_root, dir, NULL, &new_gen, NULL, NULL, - NULL, NULL); + ret = get_inode_gen(sctx->send_root, dir, &new_gen); if (ret) return ret; - ret = get_inode_info(sctx->parent_root, dir, NULL, &orig_gen, NULL, - NULL, NULL, NULL); + ret = get_inode_gen(sctx->parent_root, dir, &orig_gen); if (ret) return ret; @@ -6373,7 +7113,7 @@ static int compare_refs(struct send_ctx *sctx, struct btrfs_path *path, } leaf = path->nodes[0]; - item_size = btrfs_item_size_nr(leaf, path->slots[0]); + item_size = btrfs_item_size(leaf, path->slots[0]); ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); while (cur_offset < item_size) { extref = (struct btrfs_inode_extref *)(ptr + @@ -6400,10 +7140,53 @@ static int changed_cb(struct btrfs_path *left_path, struct btrfs_path *right_path, struct btrfs_key *key, enum btrfs_compare_tree_result result, - void *ctx) + struct send_ctx *sctx) { - int ret = 0; - struct send_ctx *sctx = ctx; + int ret; + + /* + * We can not hold the commit root semaphore here. This is because in + * the case of sending and receiving to the same filesystem, using a + * pipe, could result in a deadlock: + * + * 1) The task running send blocks on the pipe because it's full; + * + * 2) The task running receive, which is the only consumer of the pipe, + * is waiting for a transaction commit (for example due to a space + * reservation when doing a write or triggering a transaction commit + * when creating a subvolume); + * + * 3) The transaction is waiting to write lock the commit root semaphore, + * but can not acquire it since it's being held at 1). + * + * Down this call chain we write to the pipe through kernel_write(). + * The same type of problem can also happen when sending to a file that + * is stored in the same filesystem - when reserving space for a write + * into the file, we can trigger a transaction commit. + * + * Our caller has supplied us with clones of leaves from the send and + * parent roots, so we're safe here from a concurrent relocation and + * further reallocation of metadata extents while we are here. Below we + * also assert that the leaves are clones. + */ + lockdep_assert_not_held(&sctx->send_root->fs_info->commit_root_sem); + + /* + * We always have a send root, so left_path is never NULL. We will not + * have a leaf when we have reached the end of the send root but have + * not yet reached the end of the parent root. + */ + if (left_path->nodes[0]) + ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED, + &left_path->nodes[0]->bflags)); + /* + * When doing a full send we don't have a parent root, so right_path is + * NULL. When doing an incremental send, we may have reached the end of + * the parent root already, so we don't have a leaf at right_path. + */ + if (right_path && right_path->nodes[0]) + ASSERT(test_bit(EXTENT_BUFFER_UNMAPPED, + &right_path->nodes[0]->bflags)); if (result == BTRFS_COMPARE_TREE_SAME) { if (key->type == BTRFS_INODE_REF_KEY || @@ -6419,7 +7202,6 @@ static int changed_cb(struct btrfs_path *left_path, return 0; } result = BTRFS_COMPARE_TREE_CHANGED; - ret = 0; } sctx->left_path = left_path; @@ -6445,48 +7227,108 @@ static int changed_cb(struct btrfs_path *left_path, ret = changed_xattr(sctx, result); else if (key->type == BTRFS_EXTENT_DATA_KEY) ret = changed_extent(sctx, result); + else if (key->type == BTRFS_VERITY_DESC_ITEM_KEY && + key->offset == 0) + ret = changed_verity(sctx, result); } out: return ret; } +static int search_key_again(const struct send_ctx *sctx, + struct btrfs_root *root, + struct btrfs_path *path, + const struct btrfs_key *key) +{ + int ret; + + if (!path->need_commit_sem) + lockdep_assert_held_read(&root->fs_info->commit_root_sem); + + /* + * Roots used for send operations are readonly and no one can add, + * update or remove keys from them, so we should be able to find our + * key again. The only exception is deduplication, which can operate on + * readonly roots and add, update or remove keys to/from them - but at + * the moment we don't allow it to run in parallel with send. + */ + ret = btrfs_search_slot(NULL, root, key, path, 0, 0); + ASSERT(ret <= 0); + if (unlikely(ret > 0)) { + btrfs_print_tree(path->nodes[path->lowest_level], false); + btrfs_err(root->fs_info, +"send: key " BTRFS_KEY_FMT" not found in %s root %llu, lowest_level %d, slot %d", + BTRFS_KEY_FMT_VALUE(key), + (root == sctx->parent_root ? "parent" : "send"), + btrfs_root_id(root), path->lowest_level, + path->slots[path->lowest_level]); + return -EUCLEAN; + } + + return ret; +} + static int full_send_tree(struct send_ctx *sctx) { int ret; struct btrfs_root *send_root = sctx->send_root; struct btrfs_key key; - struct btrfs_path *path; - struct extent_buffer *eb; - int slot; + struct btrfs_fs_info *fs_info = send_root->fs_info; + BTRFS_PATH_AUTO_FREE(path); path = alloc_path_for_send(); if (!path) return -ENOMEM; + path->reada = READA_FORWARD_ALWAYS; key.objectid = BTRFS_FIRST_FREE_OBJECTID; key.type = BTRFS_INODE_ITEM_KEY; key.offset = 0; + down_read(&fs_info->commit_root_sem); + sctx->last_reloc_trans = fs_info->last_reloc_trans; + up_read(&fs_info->commit_root_sem); + ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0); if (ret < 0) - goto out; + return ret; if (ret) goto out_finish; while (1) { - eb = path->nodes[0]; - slot = path->slots[0]; - btrfs_item_key_to_cpu(eb, &key, slot); + btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); ret = changed_cb(path, NULL, &key, BTRFS_COMPARE_TREE_NEW, sctx); if (ret < 0) - goto out; + return ret; + + down_read(&fs_info->commit_root_sem); + if (fs_info->last_reloc_trans > sctx->last_reloc_trans) { + sctx->last_reloc_trans = fs_info->last_reloc_trans; + up_read(&fs_info->commit_root_sem); + /* + * A transaction used for relocating a block group was + * committed or is about to finish its commit. Release + * our path (leaf) and restart the search, so that we + * avoid operating on any file extent items that are + * stale, with a disk_bytenr that reflects a pre + * relocation value. This way we avoid as much as + * possible to fallback to regular writes when checking + * if we can clone file ranges. + */ + btrfs_release_path(path); + ret = search_key_again(sctx, send_root, path, &key); + if (ret < 0) + return ret; + } else { + up_read(&fs_info->commit_root_sem); + } ret = btrfs_next_item(send_root, path); if (ret < 0) - goto out; + return ret; if (ret) { ret = 0; break; @@ -6494,10 +7336,536 @@ static int full_send_tree(struct send_ctx *sctx) } out_finish: - ret = finish_inode_if_needed(sctx, 1); + return finish_inode_if_needed(sctx, 1); +} + +static int replace_node_with_clone(struct btrfs_path *path, int level) +{ + struct extent_buffer *clone; + + clone = btrfs_clone_extent_buffer(path->nodes[level]); + if (!clone) + return -ENOMEM; + + free_extent_buffer(path->nodes[level]); + path->nodes[level] = clone; + + return 0; +} + +static int tree_move_down(struct btrfs_path *path, int *level, u64 reada_min_gen) +{ + struct extent_buffer *eb; + struct extent_buffer *parent = path->nodes[*level]; + int slot = path->slots[*level]; + const int nritems = btrfs_header_nritems(parent); + u64 reada_max; + u64 reada_done = 0; + + lockdep_assert_held_read(&parent->fs_info->commit_root_sem); + ASSERT(*level != 0); + eb = btrfs_read_node_slot(parent, slot); + if (IS_ERR(eb)) + return PTR_ERR(eb); + + /* + * Trigger readahead for the next leaves we will process, so that it is + * very likely that when we need them they are already in memory and we + * will not block on disk IO. For nodes we only do readahead for one, + * since the time window between processing nodes is typically larger. + */ + reada_max = (*level == 1 ? SZ_128K : eb->fs_info->nodesize); + + for (slot++; slot < nritems && reada_done < reada_max; slot++) { + if (btrfs_node_ptr_generation(parent, slot) > reada_min_gen) { + btrfs_readahead_node_child(parent, slot); + reada_done += eb->fs_info->nodesize; + } + } + + path->nodes[*level - 1] = eb; + path->slots[*level - 1] = 0; + (*level)--; + + if (*level == 0) + return replace_node_with_clone(path, 0); + + return 0; +} + +static int tree_move_next_or_upnext(struct btrfs_path *path, + int *level, int root_level) +{ + int ret = 0; + int nritems; + nritems = btrfs_header_nritems(path->nodes[*level]); + + path->slots[*level]++; + + while (path->slots[*level] >= nritems) { + if (*level == root_level) { + path->slots[*level] = nritems - 1; + return -1; + } + + /* move upnext */ + path->slots[*level] = 0; + free_extent_buffer(path->nodes[*level]); + path->nodes[*level] = NULL; + (*level)++; + path->slots[*level]++; + + nritems = btrfs_header_nritems(path->nodes[*level]); + ret = 1; + } + return ret; +} + +/* + * Returns 1 if it had to move up and next. 0 is returned if it moved only next + * or down. + */ +static int tree_advance(struct btrfs_path *path, + int *level, int root_level, + int allow_down, + struct btrfs_key *key, + u64 reada_min_gen) +{ + int ret; + + if (*level == 0 || !allow_down) { + ret = tree_move_next_or_upnext(path, level, root_level); + } else { + ret = tree_move_down(path, level, reada_min_gen); + } + + /* + * Even if we have reached the end of a tree, ret is -1, update the key + * anyway, so that in case we need to restart due to a block group + * relocation, we can assert that the last key of the root node still + * exists in the tree. + */ + if (*level == 0) + btrfs_item_key_to_cpu(path->nodes[*level], key, + path->slots[*level]); + else + btrfs_node_key_to_cpu(path->nodes[*level], key, + path->slots[*level]); + + return ret; +} + +static int tree_compare_item(struct btrfs_path *left_path, + struct btrfs_path *right_path, + char *tmp_buf) +{ + int cmp; + int len1, len2; + unsigned long off1, off2; + + len1 = btrfs_item_size(left_path->nodes[0], left_path->slots[0]); + len2 = btrfs_item_size(right_path->nodes[0], right_path->slots[0]); + if (len1 != len2) + return 1; + + off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); + off2 = btrfs_item_ptr_offset(right_path->nodes[0], + right_path->slots[0]); + + read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); + + cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); + if (cmp) + return 1; + return 0; +} + +/* + * A transaction used for relocating a block group was committed or is about to + * finish its commit. Release our paths and restart the search, so that we are + * not using stale extent buffers: + * + * 1) For levels > 0, we are only holding references of extent buffers, without + * any locks on them, which does not prevent them from having been relocated + * and reallocated after the last time we released the commit root semaphore. + * The exception are the root nodes, for which we always have a clone, see + * the comment at btrfs_compare_trees(); + * + * 2) For leaves, level 0, we are holding copies (clones) of extent buffers, so + * we are safe from the concurrent relocation and reallocation. However they + * can have file extent items with a pre relocation disk_bytenr value, so we + * restart the start from the current commit roots and clone the new leaves so + * that we get the post relocation disk_bytenr values. Not doing so, could + * make us clone the wrong data in case there are new extents using the old + * disk_bytenr that happen to be shared. + */ +static int restart_after_relocation(struct btrfs_path *left_path, + struct btrfs_path *right_path, + const struct btrfs_key *left_key, + const struct btrfs_key *right_key, + int left_level, + int right_level, + const struct send_ctx *sctx) +{ + int root_level; + int ret; + + lockdep_assert_held_read(&sctx->send_root->fs_info->commit_root_sem); + + btrfs_release_path(left_path); + btrfs_release_path(right_path); + + /* + * Since keys can not be added or removed to/from our roots because they + * are readonly and we do not allow deduplication to run in parallel + * (which can add, remove or change keys), the layout of the trees should + * not change. + */ + left_path->lowest_level = left_level; + ret = search_key_again(sctx, sctx->send_root, left_path, left_key); + if (ret < 0) + return ret; + + right_path->lowest_level = right_level; + ret = search_key_again(sctx, sctx->parent_root, right_path, right_key); + if (ret < 0) + return ret; + + /* + * If the lowest level nodes are leaves, clone them so that they can be + * safely used by changed_cb() while not under the protection of the + * commit root semaphore, even if relocation and reallocation happens in + * parallel. + */ + if (left_level == 0) { + ret = replace_node_with_clone(left_path, 0); + if (ret < 0) + return ret; + } + + if (right_level == 0) { + ret = replace_node_with_clone(right_path, 0); + if (ret < 0) + return ret; + } + + /* + * Now clone the root nodes (unless they happen to be the leaves we have + * already cloned). This is to protect against concurrent snapshotting of + * the send and parent roots (see the comment at btrfs_compare_trees()). + */ + root_level = btrfs_header_level(sctx->send_root->commit_root); + if (root_level > 0) { + ret = replace_node_with_clone(left_path, root_level); + if (ret < 0) + return ret; + } + + root_level = btrfs_header_level(sctx->parent_root->commit_root); + if (root_level > 0) { + ret = replace_node_with_clone(right_path, root_level); + if (ret < 0) + return ret; + } + + return 0; +} + +/* + * This function compares two trees and calls the provided callback for + * every changed/new/deleted item it finds. + * If shared tree blocks are encountered, whole subtrees are skipped, making + * the compare pretty fast on snapshotted subvolumes. + * + * This currently works on commit roots only. As commit roots are read only, + * we don't do any locking. The commit roots are protected with transactions. + * Transactions are ended and rejoined when a commit is tried in between. + * + * This function checks for modifications done to the trees while comparing. + * If it detects a change, it aborts immediately. + */ +static int btrfs_compare_trees(struct btrfs_root *left_root, + struct btrfs_root *right_root, struct send_ctx *sctx) +{ + struct btrfs_fs_info *fs_info = left_root->fs_info; + int ret; + int cmp; + BTRFS_PATH_AUTO_FREE(left_path); + BTRFS_PATH_AUTO_FREE(right_path); + struct btrfs_key left_key; + struct btrfs_key right_key; + char *tmp_buf = NULL; + int left_root_level; + int right_root_level; + int left_level; + int right_level; + int left_end_reached = 0; + int right_end_reached = 0; + int advance_left = 0; + int advance_right = 0; + u64 left_blockptr; + u64 right_blockptr; + u64 left_gen; + u64 right_gen; + u64 reada_min_gen; + + left_path = btrfs_alloc_path(); + if (!left_path) { + ret = -ENOMEM; + goto out; + } + right_path = btrfs_alloc_path(); + if (!right_path) { + ret = -ENOMEM; + goto out; + } + + tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); + if (!tmp_buf) { + ret = -ENOMEM; + goto out; + } + + left_path->search_commit_root = true; + left_path->skip_locking = true; + right_path->search_commit_root = true; + right_path->skip_locking = true; + + /* + * Strategy: Go to the first items of both trees. Then do + * + * If both trees are at level 0 + * Compare keys of current items + * If left < right treat left item as new, advance left tree + * and repeat + * If left > right treat right item as deleted, advance right tree + * and repeat + * If left == right do deep compare of items, treat as changed if + * needed, advance both trees and repeat + * If both trees are at the same level but not at level 0 + * Compare keys of current nodes/leafs + * If left < right advance left tree and repeat + * If left > right advance right tree and repeat + * If left == right compare blockptrs of the next nodes/leafs + * If they match advance both trees but stay at the same level + * and repeat + * If they don't match advance both trees while allowing to go + * deeper and repeat + * If tree levels are different + * Advance the tree that needs it and repeat + * + * Advancing a tree means: + * If we are at level 0, try to go to the next slot. If that's not + * possible, go one level up and repeat. Stop when we found a level + * where we could go to the next slot. We may at this point be on a + * node or a leaf. + * + * If we are not at level 0 and not on shared tree blocks, go one + * level deeper. + * + * If we are not at level 0 and on shared tree blocks, go one slot to + * the right if possible or go up and right. + */ + + down_read(&fs_info->commit_root_sem); + left_level = btrfs_header_level(left_root->commit_root); + left_root_level = left_level; + /* + * We clone the root node of the send and parent roots to prevent races + * with snapshot creation of these roots. Snapshot creation COWs the + * root node of a tree, so after the transaction is committed the old + * extent can be reallocated while this send operation is still ongoing. + * So we clone them, under the commit root semaphore, to be race free. + */ + left_path->nodes[left_level] = + btrfs_clone_extent_buffer(left_root->commit_root); + if (!left_path->nodes[left_level]) { + ret = -ENOMEM; + goto out_unlock; + } + + right_level = btrfs_header_level(right_root->commit_root); + right_root_level = right_level; + right_path->nodes[right_level] = + btrfs_clone_extent_buffer(right_root->commit_root); + if (!right_path->nodes[right_level]) { + ret = -ENOMEM; + goto out_unlock; + } + /* + * Our right root is the parent root, while the left root is the "send" + * root. We know that all new nodes/leaves in the left root must have + * a generation greater than the right root's generation, so we trigger + * readahead for those nodes and leaves of the left root, as we know we + * will need to read them at some point. + */ + reada_min_gen = btrfs_header_generation(right_root->commit_root); + + if (left_level == 0) + btrfs_item_key_to_cpu(left_path->nodes[left_level], + &left_key, left_path->slots[left_level]); + else + btrfs_node_key_to_cpu(left_path->nodes[left_level], + &left_key, left_path->slots[left_level]); + if (right_level == 0) + btrfs_item_key_to_cpu(right_path->nodes[right_level], + &right_key, right_path->slots[right_level]); + else + btrfs_node_key_to_cpu(right_path->nodes[right_level], + &right_key, right_path->slots[right_level]); + + sctx->last_reloc_trans = fs_info->last_reloc_trans; + + while (1) { + if (need_resched() || + rwsem_is_contended(&fs_info->commit_root_sem)) { + up_read(&fs_info->commit_root_sem); + cond_resched(); + down_read(&fs_info->commit_root_sem); + } + + if (fs_info->last_reloc_trans > sctx->last_reloc_trans) { + ret = restart_after_relocation(left_path, right_path, + &left_key, &right_key, + left_level, right_level, + sctx); + if (ret < 0) + goto out_unlock; + sctx->last_reloc_trans = fs_info->last_reloc_trans; + } + + if (advance_left && !left_end_reached) { + ret = tree_advance(left_path, &left_level, + left_root_level, + advance_left != ADVANCE_ONLY_NEXT, + &left_key, reada_min_gen); + if (ret == -1) + left_end_reached = ADVANCE; + else if (ret < 0) + goto out_unlock; + advance_left = 0; + } + if (advance_right && !right_end_reached) { + ret = tree_advance(right_path, &right_level, + right_root_level, + advance_right != ADVANCE_ONLY_NEXT, + &right_key, reada_min_gen); + if (ret == -1) + right_end_reached = ADVANCE; + else if (ret < 0) + goto out_unlock; + advance_right = 0; + } + + if (left_end_reached && right_end_reached) { + ret = 0; + goto out_unlock; + } else if (left_end_reached) { + if (right_level == 0) { + up_read(&fs_info->commit_root_sem); + ret = changed_cb(left_path, right_path, + &right_key, + BTRFS_COMPARE_TREE_DELETED, + sctx); + if (ret < 0) + goto out; + down_read(&fs_info->commit_root_sem); + } + advance_right = ADVANCE; + continue; + } else if (right_end_reached) { + if (left_level == 0) { + up_read(&fs_info->commit_root_sem); + ret = changed_cb(left_path, right_path, + &left_key, + BTRFS_COMPARE_TREE_NEW, + sctx); + if (ret < 0) + goto out; + down_read(&fs_info->commit_root_sem); + } + advance_left = ADVANCE; + continue; + } + + if (left_level == 0 && right_level == 0) { + up_read(&fs_info->commit_root_sem); + cmp = btrfs_comp_cpu_keys(&left_key, &right_key); + if (cmp < 0) { + ret = changed_cb(left_path, right_path, + &left_key, + BTRFS_COMPARE_TREE_NEW, + sctx); + advance_left = ADVANCE; + } else if (cmp > 0) { + ret = changed_cb(left_path, right_path, + &right_key, + BTRFS_COMPARE_TREE_DELETED, + sctx); + advance_right = ADVANCE; + } else { + enum btrfs_compare_tree_result result; + + WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); + ret = tree_compare_item(left_path, right_path, + tmp_buf); + if (ret) + result = BTRFS_COMPARE_TREE_CHANGED; + else + result = BTRFS_COMPARE_TREE_SAME; + ret = changed_cb(left_path, right_path, + &left_key, result, sctx); + advance_left = ADVANCE; + advance_right = ADVANCE; + } + + if (ret < 0) + goto out; + down_read(&fs_info->commit_root_sem); + } else if (left_level == right_level) { + cmp = btrfs_comp_cpu_keys(&left_key, &right_key); + if (cmp < 0) { + advance_left = ADVANCE; + } else if (cmp > 0) { + advance_right = ADVANCE; + } else { + left_blockptr = btrfs_node_blockptr( + left_path->nodes[left_level], + left_path->slots[left_level]); + right_blockptr = btrfs_node_blockptr( + right_path->nodes[right_level], + right_path->slots[right_level]); + left_gen = btrfs_node_ptr_generation( + left_path->nodes[left_level], + left_path->slots[left_level]); + right_gen = btrfs_node_ptr_generation( + right_path->nodes[right_level], + right_path->slots[right_level]); + if (left_blockptr == right_blockptr && + left_gen == right_gen) { + /* + * As we're on a shared block, don't + * allow to go deeper. + */ + advance_left = ADVANCE_ONLY_NEXT; + advance_right = ADVANCE_ONLY_NEXT; + } else { + advance_left = ADVANCE; + advance_right = ADVANCE; + } + } + } else if (left_level < right_level) { + advance_right = ADVANCE; + } else { + advance_left = ADVANCE; + } + } + +out_unlock: + up_read(&fs_info->commit_root_sem); out: - btrfs_free_path(path); + kvfree(tmp_buf); return ret; } @@ -6516,8 +7884,7 @@ static int send_subvol(struct send_ctx *sctx) goto out; if (sctx->parent_root) { - ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, - changed_cb, sctx); + ret = btrfs_compare_trees(sctx->send_root, sctx->parent_root, sctx); if (ret < 0) goto out; ret = finish_inode_if_needed(sctx, 1); @@ -6549,34 +7916,50 @@ out: */ static int ensure_commit_roots_uptodate(struct send_ctx *sctx) { - int i; - struct btrfs_trans_handle *trans = NULL; - -again: - if (sctx->parent_root && - sctx->parent_root->node != sctx->parent_root->commit_root) - goto commit_trans; + struct btrfs_root *root = sctx->parent_root; - for (i = 0; i < sctx->clone_roots_cnt; i++) - if (sctx->clone_roots[i].root->node != - sctx->clone_roots[i].root->commit_root) - goto commit_trans; + if (root && root->node != root->commit_root) + return btrfs_commit_current_transaction(root); - if (trans) - return btrfs_end_transaction(trans); + for (int i = 0; i < sctx->clone_roots_cnt; i++) { + root = sctx->clone_roots[i].root; + if (root->node != root->commit_root) + return btrfs_commit_current_transaction(root); + } return 0; +} -commit_trans: - /* Use any root, all fs roots will get their commit roots updated. */ - if (!trans) { - trans = btrfs_join_transaction(sctx->send_root); - if (IS_ERR(trans)) - return PTR_ERR(trans); - goto again; +/* + * Make sure any existing delalloc is flushed for any root used by a send + * operation so that we do not miss any data and we do not race with writeback + * finishing and changing a tree while send is using the tree. This could + * happen if a subvolume is in RW mode, has delalloc, is turned to RO mode and + * a send operation then uses the subvolume. + * After flushing delalloc ensure_commit_roots_uptodate() must be called. + */ +static int flush_delalloc_roots(struct send_ctx *sctx) +{ + struct btrfs_root *root = sctx->parent_root; + int ret; + int i; + + if (root) { + ret = btrfs_start_delalloc_snapshot(root, false); + if (ret) + return ret; + btrfs_wait_ordered_extents(root, U64_MAX, NULL); } - return btrfs_commit_transaction(trans); + for (i = 0; i < sctx->clone_roots_cnt; i++) { + root = sctx->clone_roots[i].root; + ret = btrfs_start_delalloc_snapshot(root, false); + if (ret) + return ret; + btrfs_wait_ordered_extents(root, U64_MAX, NULL); + } + + return 0; } static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) @@ -6590,24 +7973,30 @@ static void btrfs_root_dec_send_in_progress(struct btrfs_root* root) if (root->send_in_progress < 0) btrfs_err(root->fs_info, "send_in_progress unbalanced %d root %llu", - root->send_in_progress, root->root_key.objectid); + root->send_in_progress, btrfs_root_id(root)); spin_unlock(&root->root_item_lock); } -long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) +static void dedupe_in_progress_warn(const struct btrfs_root *root) +{ + btrfs_warn_rl(root->fs_info, +"cannot use root %llu for send while deduplications on it are in progress (%d in progress)", + btrfs_root_id(root), root->dedupe_in_progress); +} + +long btrfs_ioctl_send(struct btrfs_root *send_root, const struct btrfs_ioctl_send_args *arg) { int ret = 0; - struct btrfs_root *send_root = BTRFS_I(file_inode(mnt_file))->root; struct btrfs_fs_info *fs_info = send_root->fs_info; struct btrfs_root *clone_root; - struct btrfs_key key; struct send_ctx *sctx = NULL; u32 i; u64 *clone_sources_tmp = NULL; int clone_sources_to_rollback = 0; - unsigned alloc_size; + size_t alloc_size; int sort_clone_roots = 0; - int index; + struct btrfs_lru_cache_entry *entry; + struct btrfs_lru_cache_entry *tmp; if (!capable(CAP_SYS_ADMIN)) return -EPERM; @@ -6617,44 +8006,40 @@ long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) * making it RW. This also protects against deletion. */ spin_lock(&send_root->root_item_lock); - send_root->send_in_progress++; - spin_unlock(&send_root->root_item_lock); - /* - * This is done when we lookup the root, it should already be complete - * by the time we get here. - */ - WARN_ON(send_root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE); - - /* - * Userspace tools do the checks and warn the user if it's - * not RO. + * Unlikely but possible, if the subvolume is marked for deletion but + * is slow to remove the directory entry, send can still be started. */ + if (btrfs_root_dead(send_root)) { + spin_unlock(&send_root->root_item_lock); + return -EPERM; + } + /* Userspace tools do the checks and warn the user if it's not RO. */ if (!btrfs_root_readonly(send_root)) { - ret = -EPERM; - goto out; + spin_unlock(&send_root->root_item_lock); + return -EPERM; } + if (send_root->dedupe_in_progress) { + dedupe_in_progress_warn(send_root); + spin_unlock(&send_root->root_item_lock); + return -EAGAIN; + } + send_root->send_in_progress++; + spin_unlock(&send_root->root_item_lock); /* * Check that we don't overflow at later allocations, we request * clone_sources_count + 1 items, and compare to unsigned long inside - * access_ok. + * access_ok. Also set an upper limit for allocation size so this can't + * easily exhaust memory. Max number of clone sources is about 200K. */ - if (arg->clone_sources_count > - ULONG_MAX / sizeof(struct clone_root) - 1) { + if (arg->clone_sources_count > SZ_8M / sizeof(struct clone_root)) { ret = -EINVAL; goto out; } - if (!access_ok(arg->clone_sources, - sizeof(*arg->clone_sources) * - arg->clone_sources_count)) { - ret = -EFAULT; - goto out; - } - if (arg->flags & ~BTRFS_SEND_FLAG_MASK) { - ret = -EINVAL; + ret = -EOPNOTSUPP; goto out; } @@ -6664,57 +8049,92 @@ long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) goto out; } + init_path(&sctx->cur_inode_path); INIT_LIST_HEAD(&sctx->new_refs); INIT_LIST_HEAD(&sctx->deleted_refs); - INIT_RADIX_TREE(&sctx->name_cache, GFP_KERNEL); - INIT_LIST_HEAD(&sctx->name_cache_list); + + btrfs_lru_cache_init(&sctx->name_cache, SEND_MAX_NAME_CACHE_SIZE); + btrfs_lru_cache_init(&sctx->backref_cache, SEND_MAX_BACKREF_CACHE_SIZE); + btrfs_lru_cache_init(&sctx->dir_created_cache, + SEND_MAX_DIR_CREATED_CACHE_SIZE); + /* + * This cache is periodically trimmed to a fixed size elsewhere, see + * cache_dir_utimes() and trim_dir_utimes_cache(). + */ + btrfs_lru_cache_init(&sctx->dir_utimes_cache, 0); + + sctx->pending_dir_moves = RB_ROOT; + sctx->waiting_dir_moves = RB_ROOT; + sctx->orphan_dirs = RB_ROOT; + sctx->rbtree_new_refs = RB_ROOT; + sctx->rbtree_deleted_refs = RB_ROOT; sctx->flags = arg->flags; - sctx->send_filp = fget(arg->send_fd); - if (!sctx->send_filp) { - ret = -EBADF; + if (arg->flags & BTRFS_SEND_FLAG_VERSION) { + if (arg->version > BTRFS_SEND_STREAM_VERSION) { + ret = -EPROTO; + goto out; + } + /* Zero means "use the highest version" */ + sctx->proto = arg->version ?: BTRFS_SEND_STREAM_VERSION; + } else { + sctx->proto = 1; + } + if ((arg->flags & BTRFS_SEND_FLAG_COMPRESSED) && sctx->proto < 2) { + ret = -EINVAL; goto out; } - sctx->send_root = send_root; - /* - * Unlikely but possible, if the subvolume is marked for deletion but - * is slow to remove the directory entry, send can still be started - */ - if (btrfs_root_dead(sctx->send_root)) { - ret = -EPERM; + sctx->send_filp = fget(arg->send_fd); + if (!sctx->send_filp || !(sctx->send_filp->f_mode & FMODE_WRITE)) { + ret = -EBADF; goto out; } + sctx->send_root = send_root; sctx->clone_roots_cnt = arg->clone_sources_count; - sctx->send_max_size = BTRFS_SEND_BUF_SIZE; - sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL); - if (!sctx->send_buf) { - ret = -ENOMEM; - goto out; - } + if (sctx->proto >= 2) { + u32 send_buf_num_pages; - sctx->read_buf = kvmalloc(BTRFS_SEND_READ_SIZE, GFP_KERNEL); - if (!sctx->read_buf) { + sctx->send_max_size = BTRFS_SEND_BUF_SIZE_V2; + sctx->send_buf = vmalloc(sctx->send_max_size); + if (!sctx->send_buf) { + ret = -ENOMEM; + goto out; + } + send_buf_num_pages = sctx->send_max_size >> PAGE_SHIFT; + sctx->send_buf_pages = kcalloc(send_buf_num_pages, + sizeof(*sctx->send_buf_pages), + GFP_KERNEL); + if (!sctx->send_buf_pages) { + ret = -ENOMEM; + goto out; + } + for (i = 0; i < send_buf_num_pages; i++) { + sctx->send_buf_pages[i] = + vmalloc_to_page(sctx->send_buf + (i << PAGE_SHIFT)); + } + } else { + sctx->send_max_size = BTRFS_SEND_BUF_SIZE_V1; + sctx->send_buf = kvmalloc(sctx->send_max_size, GFP_KERNEL); + } + if (!sctx->send_buf) { ret = -ENOMEM; goto out; } - sctx->pending_dir_moves = RB_ROOT; - sctx->waiting_dir_moves = RB_ROOT; - sctx->orphan_dirs = RB_ROOT; - - alloc_size = sizeof(struct clone_root) * (arg->clone_sources_count + 1); - - sctx->clone_roots = kzalloc(alloc_size, GFP_KERNEL); + sctx->clone_roots = kvcalloc(arg->clone_sources_count + 1, + sizeof(*sctx->clone_roots), + GFP_KERNEL); if (!sctx->clone_roots) { ret = -ENOMEM; goto out; } - alloc_size = arg->clone_sources_count * sizeof(*arg->clone_sources); + alloc_size = array_size(sizeof(*arg->clone_sources), + arg->clone_sources_count); if (arg->clone_sources_count) { clone_sources_tmp = kvmalloc(alloc_size, GFP_KERNEL); @@ -6731,15 +8151,9 @@ long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) } for (i = 0; i < arg->clone_sources_count; i++) { - key.objectid = clone_sources_tmp[i]; - key.type = BTRFS_ROOT_ITEM_KEY; - key.offset = (u64)-1; - - index = srcu_read_lock(&fs_info->subvol_srcu); - - clone_root = btrfs_read_fs_root_no_name(fs_info, &key); + clone_root = btrfs_get_fs_root(fs_info, + clone_sources_tmp[i], true); if (IS_ERR(clone_root)) { - srcu_read_unlock(&fs_info->subvol_srcu, index); ret = PTR_ERR(clone_root); goto out; } @@ -6747,13 +8161,19 @@ long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) if (!btrfs_root_readonly(clone_root) || btrfs_root_dead(clone_root)) { spin_unlock(&clone_root->root_item_lock); - srcu_read_unlock(&fs_info->subvol_srcu, index); + btrfs_put_root(clone_root); ret = -EPERM; goto out; } + if (clone_root->dedupe_in_progress) { + dedupe_in_progress_warn(clone_root); + spin_unlock(&clone_root->root_item_lock); + btrfs_put_root(clone_root); + ret = -EAGAIN; + goto out; + } clone_root->send_in_progress++; spin_unlock(&clone_root->root_item_lock); - srcu_read_unlock(&fs_info->subvol_srcu, index); sctx->clone_roots[i].root = clone_root; clone_sources_to_rollback = i + 1; @@ -6763,15 +8183,9 @@ long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) } if (arg->parent_root) { - key.objectid = arg->parent_root; - key.type = BTRFS_ROOT_ITEM_KEY; - key.offset = (u64)-1; - - index = srcu_read_lock(&fs_info->subvol_srcu); - - sctx->parent_root = btrfs_read_fs_root_no_name(fs_info, &key); + sctx->parent_root = btrfs_get_fs_root(fs_info, arg->parent_root, + true); if (IS_ERR(sctx->parent_root)) { - srcu_read_unlock(&fs_info->subvol_srcu, index); ret = PTR_ERR(sctx->parent_root); goto out; } @@ -6781,13 +8195,16 @@ long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) if (!btrfs_root_readonly(sctx->parent_root) || btrfs_root_dead(sctx->parent_root)) { spin_unlock(&sctx->parent_root->root_item_lock); - srcu_read_unlock(&fs_info->subvol_srcu, index); ret = -EPERM; goto out; } + if (sctx->parent_root->dedupe_in_progress) { + dedupe_in_progress_warn(sctx->parent_root); + spin_unlock(&sctx->parent_root->root_item_lock); + ret = -EAGAIN; + goto out; + } spin_unlock(&sctx->parent_root->root_item_lock); - - srcu_read_unlock(&fs_info->subvol_srcu, index); } /* @@ -6795,7 +8212,8 @@ long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) * is behind the current send position. This is checked while searching * for possible clone sources. */ - sctx->clone_roots[sctx->clone_roots_cnt++].root = sctx->send_root; + sctx->clone_roots[sctx->clone_roots_cnt++].root = + btrfs_grab_root(sctx->send_root); /* We do a bsearch later */ sort(sctx->clone_roots, sctx->clone_roots_cnt, @@ -6803,16 +8221,25 @@ long btrfs_ioctl_send(struct file *mnt_file, struct btrfs_ioctl_send_args *arg) NULL); sort_clone_roots = 1; + ret = flush_delalloc_roots(sctx); + if (ret) + goto out; + ret = ensure_commit_roots_uptodate(sctx); if (ret) goto out; - current->journal_info = BTRFS_SEND_TRANS_STUB; ret = send_subvol(sctx); - current->journal_info = NULL; if (ret < 0) goto out; + btrfs_lru_cache_for_each_entry_safe(&sctx->dir_utimes_cache, entry, tmp) { + ret = send_utimes(sctx, entry->key, entry->gen); + if (ret < 0) + goto out; + btrfs_lru_cache_remove(&sctx->dir_utimes_cache, entry); + } + if (!(sctx->flags & BTRFS_SEND_FLAG_OMIT_END_CMD)) { ret = begin_cmd(sctx, BTRFS_SEND_C_END); if (ret < 0) @@ -6862,18 +8289,24 @@ out: } if (sort_clone_roots) { - for (i = 0; i < sctx->clone_roots_cnt; i++) + for (i = 0; i < sctx->clone_roots_cnt; i++) { btrfs_root_dec_send_in_progress( sctx->clone_roots[i].root); + btrfs_put_root(sctx->clone_roots[i].root); + } } else { - for (i = 0; sctx && i < clone_sources_to_rollback; i++) + for (i = 0; sctx && i < clone_sources_to_rollback; i++) { btrfs_root_dec_send_in_progress( sctx->clone_roots[i].root); + btrfs_put_root(sctx->clone_roots[i].root); + } btrfs_root_dec_send_in_progress(send_root); } - if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) + if (sctx && !IS_ERR_OR_NULL(sctx->parent_root)) { btrfs_root_dec_send_in_progress(sctx->parent_root); + btrfs_put_root(sctx->parent_root); + } kvfree(clone_sources_tmp); @@ -6882,10 +8315,19 @@ out: fput(sctx->send_filp); kvfree(sctx->clone_roots); + kfree(sctx->send_buf_pages); kvfree(sctx->send_buf); - kvfree(sctx->read_buf); + kvfree(sctx->verity_descriptor); + + close_current_inode(sctx); + + btrfs_lru_cache_clear(&sctx->name_cache); + btrfs_lru_cache_clear(&sctx->backref_cache); + btrfs_lru_cache_clear(&sctx->dir_created_cache); + btrfs_lru_cache_clear(&sctx->dir_utimes_cache); - name_cache_free(sctx); + if (sctx->cur_inode_path.buf != sctx->cur_inode_path.inline_buf) + kfree(sctx->cur_inode_path.buf); kfree(sctx); } |