1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Copyright (C) 2007 Oracle. All rights reserved. 4 */ 5 6 #ifndef BTRFS_TRANSACTION_H 7 #define BTRFS_TRANSACTION_H 8 9 #include <linux/atomic.h> 10 #include <linux/build_bug.h> 11 #include <linux/refcount.h> 12 #include <linux/list.h> 13 #include <linux/time64.h> 14 #include <linux/mutex.h> 15 #include <linux/wait.h> 16 #include "btrfs_inode.h" 17 #include "delayed-ref.h" 18 19 struct dentry; 20 struct inode; 21 struct btrfs_pending_snapshot; 22 struct btrfs_fs_info; 23 struct btrfs_root_item; 24 struct btrfs_root; 25 struct btrfs_path; 26 struct extent_buffer; 27 28 /* 29 * Signal that a direct IO write is in progress, to avoid deadlock for sync 30 * direct IO writes when fsync is called during the direct IO write path. 31 */ 32 #define BTRFS_TRANS_DIO_WRITE_STUB ((void *) 1) 33 34 /* Radix-tree tag for roots that are part of the transaction. */ 35 #define BTRFS_ROOT_TRANS_TAG 0 36 37 enum btrfs_trans_state { 38 TRANS_STATE_RUNNING, 39 TRANS_STATE_COMMIT_PREP, 40 TRANS_STATE_COMMIT_START, 41 TRANS_STATE_COMMIT_DOING, 42 TRANS_STATE_UNBLOCKED, 43 TRANS_STATE_SUPER_COMMITTED, 44 TRANS_STATE_COMPLETED, 45 TRANS_STATE_MAX, 46 }; 47 48 #define BTRFS_TRANS_HAVE_FREE_BGS 0 49 #define BTRFS_TRANS_DIRTY_BG_RUN 1 50 51 struct btrfs_transaction { 52 u64 transid; 53 /* 54 * total external writers(USERSPACE/START/ATTACH) in this 55 * transaction, it must be zero before the transaction is 56 * being committed 57 */ 58 atomic_t num_extwriters; 59 /* 60 * total writers in this transaction, it must be zero before the 61 * transaction can end 62 */ 63 atomic_t num_writers; 64 refcount_t use_count; 65 66 unsigned long flags; 67 68 /* Be protected by fs_info->trans_lock when we want to change it. */ 69 enum btrfs_trans_state state; 70 int aborted; 71 struct list_head list; 72 struct extent_io_tree dirty_pages; 73 time64_t start_time; 74 wait_queue_head_t writer_wait; 75 wait_queue_head_t commit_wait; 76 struct list_head pending_snapshots; 77 struct list_head dev_update_list; 78 struct list_head switch_commits; 79 struct list_head dirty_bgs; 80 struct list_head dropped_roots; 81 struct extent_io_tree pinned_extents; 82 83 /* 84 * We need to make sure block group deletion doesn't race with the 85 * dirty block group item updates done outside the commit critical 86 * section. This mutex keeps them from stomping on each other. 87 */ 88 struct mutex dirty_bgs_update_mutex; 89 spinlock_t dirty_bgs_lock; 90 /* Protected by spin lock fs_info->unused_bgs_lock. */ 91 struct list_head deleted_bgs; 92 spinlock_t dropped_roots_lock; 93 struct btrfs_delayed_ref_root delayed_refs; 94 struct btrfs_fs_info *fs_info; 95 96 /* 97 * Number of ordered extents the transaction must wait for before 98 * committing. These are ordered extents started by a fast fsync. 99 */ 100 atomic_t pending_ordered; 101 wait_queue_head_t pending_wait; 102 }; 103 104 /* 105 * Number of extent buffers a transaction handle tracks for writeback 106 * inhibition. The CLOCK reference bits pack into a u32 so this must not exceed 107 * 32, and keeping it a power of two lets the compiler reduce the CLOCK hand 108 * modulo to a mask. 109 */ 110 #define BTRFS_INHIBITED_EBS_SLOTS 8 111 112 static_assert(BTRFS_INHIBITED_EBS_SLOTS <= 32); 113 static_assert(BTRFS_INHIBITED_EBS_SLOTS != 0 && 114 (BTRFS_INHIBITED_EBS_SLOTS & (BTRFS_INHIBITED_EBS_SLOTS - 1)) == 0); 115 116 struct btrfs_trans_handle { 117 u64 transid; 118 u64 bytes_reserved; 119 u64 delayed_refs_bytes_reserved; 120 u64 chunk_bytes_reserved; 121 unsigned long delayed_ref_updates; 122 unsigned long delayed_ref_csum_deletions; 123 struct btrfs_transaction *transaction; 124 struct btrfs_block_rsv *block_rsv; 125 struct btrfs_block_rsv *orig_rsv; 126 /* Set by a task that wants to create a snapshot. */ 127 struct btrfs_pending_snapshot *pending_snapshot; 128 refcount_t use_count; 129 unsigned int type; 130 /* 131 * Error code of transaction abort, set outside of locks and must use 132 * the READ_ONCE/WRITE_ONCE access 133 */ 134 short aborted; 135 bool adding_csums; 136 bool allocating_chunk; 137 bool removing_chunk; 138 bool reloc_reserved; 139 bool in_fsync; 140 struct btrfs_fs_info *fs_info; 141 struct list_head new_bgs; 142 struct btrfs_block_rsv delayed_rsv; 143 144 /* Extent buffers this handle has inhibited writeback on. */ 145 struct extent_buffer *inhibited_ebs[BTRFS_INHIBITED_EBS_SLOTS]; 146 /* CLOCK reference bit per slot. */ 147 u32 inhibited_ebs_referenced; 148 u32 nr_inhibited_ebs; 149 /* CLOCK hand. */ 150 u32 inhibited_ebs_hand; 151 }; 152 153 /* 154 * The abort status can be changed between calls and is not protected by locks. 155 * This accepts btrfs_transaction and btrfs_trans_handle as types. Once it's 156 * set to a non-zero value it does not change, so the macro should be in checks 157 * but is not necessary for further reads of the value. 158 */ 159 #define TRANS_ABORTED(trans) (unlikely(READ_ONCE((trans)->aborted))) 160 161 struct btrfs_pending_snapshot { 162 struct dentry *dentry; 163 struct btrfs_inode *dir; 164 struct btrfs_root *root; 165 struct btrfs_root_item *root_item; 166 struct btrfs_root *snap; 167 struct btrfs_qgroup_inherit *inherit; 168 struct btrfs_path *path; 169 /* block reservation for the operation */ 170 struct btrfs_block_rsv block_rsv; 171 /* extra metadata reservation for relocation */ 172 int error; 173 /* Preallocated anonymous block device number */ 174 dev_t anon_dev; 175 bool readonly; 176 struct list_head list; 177 }; 178 179 static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans, 180 struct btrfs_inode *inode) 181 { 182 spin_lock(&inode->lock); 183 inode->last_trans = trans->transaction->transid; 184 inode->last_sub_trans = btrfs_get_root_log_transid(inode->root); 185 inode->last_log_commit = inode->last_sub_trans - 1; 186 spin_unlock(&inode->lock); 187 } 188 189 /* 190 * Make qgroup codes to skip given qgroupid, means the old/new_roots for 191 * qgroup won't contain the qgroupid in it. 192 */ 193 static inline void btrfs_set_skip_qgroup(struct btrfs_trans_handle *trans, 194 u64 qgroupid) 195 { 196 struct btrfs_delayed_ref_root *delayed_refs; 197 198 delayed_refs = &trans->transaction->delayed_refs; 199 WARN_ON(delayed_refs->qgroup_to_skip); 200 delayed_refs->qgroup_to_skip = qgroupid; 201 } 202 203 static inline void btrfs_clear_skip_qgroup(struct btrfs_trans_handle *trans) 204 { 205 struct btrfs_delayed_ref_root *delayed_refs; 206 207 delayed_refs = &trans->transaction->delayed_refs; 208 WARN_ON(!delayed_refs->qgroup_to_skip); 209 delayed_refs->qgroup_to_skip = 0; 210 } 211 212 /* 213 * We want the transaction abort to print stack trace only for errors where the 214 * cause could be a bug, eg. due to ENOSPC, and not for common errors that are 215 * caused by external factors. 216 */ 217 static inline bool btrfs_abort_should_print_stack(int error) 218 { 219 switch (error) { 220 case -EIO: 221 case -EROFS: 222 case -ENOMEM: 223 return false; 224 } 225 return true; 226 } 227 228 /* 229 * Compile-time and run-time verification of error passed to transaction abort. 230 * Direct constants will be caught at compile time, errors read from variables 231 * can be caught only at run-time and will warn under debugging config. 232 * 233 * How verification works: 234 * - accepted builtin constants are all -EIO and such 235 * - for compile-time check, invalid condition produces a negative-sized array 236 * type, valid zero-sized 237 * - when a variable is passed as error the first check is a no-op 238 * - with enabled debugging, the second array type size is constructed from the 239 * real variable value, valid condition produces array of size 1 240 * - sizeof(type) does not generate any code 241 */ 242 #define VERIFY_NEGATIVE_ERROR(error) \ 243 do { \ 244 (void)sizeof(char[-!(__builtin_constant_p(error) ? (error) < 0 : 1)]); \ 245 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { \ 246 if (sizeof(char[(error) < 0]) != 1) \ 247 DEBUG_WARN("error >= 0 passed to btrfs_abort_transaction()"); \ 248 } \ 249 } while(0) 250 251 /* 252 * Call btrfs_abort_transaction() as early as possible when an error condition 253 * is detected, that way the exact stack trace is reported for some errors. 254 * 255 * Error number must be negative as it encodes whether it's the first abort. 256 */ 257 #define btrfs_abort_transaction(trans, error) \ 258 do { \ 259 int __error = (error); \ 260 \ 261 VERIFY_NEGATIVE_ERROR(error); \ 262 /* Report first abort since mount */ \ 263 if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED, \ 264 &((trans)->fs_info->fs_state))) { \ 265 WARN_ON(btrfs_abort_should_print_stack(__error)); \ 266 __error = -__error; \ 267 } \ 268 __btrfs_abort_transaction((trans), __func__, \ 269 __LINE__, __error); \ 270 } while (0) 271 272 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 273 bool btrfs_is_dummy_transaction(const struct btrfs_trans_handle *trans); 274 void btrfs_init_dummy_trans(struct btrfs_trans_handle *trans, 275 struct btrfs_fs_info *fs_info); 276 #else 277 static inline bool btrfs_is_dummy_transaction(const struct btrfs_trans_handle *trans) 278 { 279 return false; 280 } 281 #endif 282 283 int btrfs_end_transaction(struct btrfs_trans_handle *trans); 284 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 285 unsigned int num_items); 286 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 287 struct btrfs_root *root, 288 unsigned int num_items); 289 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root); 290 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root); 291 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root); 292 struct btrfs_trans_handle *btrfs_attach_transaction_barrier( 293 struct btrfs_root *root); 294 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid); 295 296 void btrfs_add_dead_root(struct btrfs_root *root); 297 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info); 298 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info); 299 int btrfs_commit_transaction(struct btrfs_trans_handle *trans); 300 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans); 301 int btrfs_commit_current_transaction(struct btrfs_root *root); 302 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans); 303 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans); 304 void btrfs_throttle(struct btrfs_fs_info *fs_info); 305 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 306 struct btrfs_root *root); 307 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 308 struct extent_io_tree *dirty_pages, int mark); 309 int btrfs_wait_tree_log_extents(struct btrfs_root *root, int mark); 310 int btrfs_transaction_blocked(struct btrfs_fs_info *info); 311 void btrfs_put_transaction(struct btrfs_transaction *transaction); 312 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 313 struct btrfs_root *root); 314 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans); 315 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans, 316 const char *function, 317 unsigned int line, int error); 318 319 int __init btrfs_transaction_init(void); 320 void __cold btrfs_transaction_exit(void); 321 322 #endif 323