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 #define BTRFS_TRANS_CACHE_ENOSPC 2 51 52 struct btrfs_transaction { 53 u64 transid; 54 /* 55 * total external writers(USERSPACE/START/ATTACH) in this 56 * transaction, it must be zero before the transaction is 57 * being committed 58 */ 59 atomic_t num_extwriters; 60 /* 61 * total writers in this transaction, it must be zero before the 62 * transaction can end 63 */ 64 atomic_t num_writers; 65 refcount_t use_count; 66 67 unsigned long flags; 68 69 /* Be protected by fs_info->trans_lock when we want to change it. */ 70 enum btrfs_trans_state state; 71 int aborted; 72 struct list_head list; 73 struct extent_io_tree dirty_pages; 74 time64_t start_time; 75 wait_queue_head_t writer_wait; 76 wait_queue_head_t commit_wait; 77 struct list_head pending_snapshots; 78 struct list_head dev_update_list; 79 struct list_head switch_commits; 80 struct list_head dirty_bgs; 81 82 /* 83 * There is no explicit lock which protects io_bgs, rather its 84 * consistency is implied by the fact that all the sites which modify 85 * it do so under some form of transaction critical section, namely: 86 * 87 * - btrfs_start_dirty_block_groups - This function can only ever be 88 * run by one of the transaction committers. Refer to 89 * BTRFS_TRANS_DIRTY_BG_RUN usage in btrfs_commit_transaction 90 * 91 * - btrfs_write_dirty_blockgroups - this is called by 92 * commit_cowonly_roots from transaction critical section 93 * (TRANS_STATE_COMMIT_DOING) 94 * 95 * - btrfs_cleanup_dirty_bgs - called on transaction abort 96 */ 97 struct list_head io_bgs; 98 struct list_head dropped_roots; 99 struct extent_io_tree pinned_extents; 100 101 /* 102 * we need to make sure block group deletion doesn't race with 103 * free space cache writeout. This mutex keeps them from stomping 104 * on each other 105 */ 106 struct mutex cache_write_mutex; 107 spinlock_t dirty_bgs_lock; 108 /* Protected by spin lock fs_info->unused_bgs_lock. */ 109 struct list_head deleted_bgs; 110 spinlock_t dropped_roots_lock; 111 struct btrfs_delayed_ref_root delayed_refs; 112 struct btrfs_fs_info *fs_info; 113 114 /* 115 * Number of ordered extents the transaction must wait for before 116 * committing. These are ordered extents started by a fast fsync. 117 */ 118 atomic_t pending_ordered; 119 wait_queue_head_t pending_wait; 120 }; 121 122 enum { 123 ENUM_BIT(__TRANS_FREEZABLE), 124 ENUM_BIT(__TRANS_START), 125 ENUM_BIT(__TRANS_ATTACH), 126 ENUM_BIT(__TRANS_JOIN), 127 ENUM_BIT(__TRANS_JOIN_NOLOCK), 128 ENUM_BIT(__TRANS_DUMMY), 129 ENUM_BIT(__TRANS_JOIN_NOSTART), 130 }; 131 132 #define TRANS_START (__TRANS_START | __TRANS_FREEZABLE) 133 #define TRANS_ATTACH (__TRANS_ATTACH) 134 #define TRANS_JOIN (__TRANS_JOIN | __TRANS_FREEZABLE) 135 #define TRANS_JOIN_NOLOCK (__TRANS_JOIN_NOLOCK) 136 #define TRANS_JOIN_NOSTART (__TRANS_JOIN_NOSTART) 137 138 #define TRANS_EXTWRITERS (__TRANS_START | __TRANS_ATTACH) 139 140 /* 141 * Number of extent buffers a transaction handle tracks for writeback 142 * inhibition. The CLOCK reference bits pack into a u32 so this must not exceed 143 * 32, and keeping it a power of two lets the compiler reduce the CLOCK hand 144 * modulo to a mask. 145 */ 146 #define BTRFS_INHIBITED_EBS_SLOTS 8 147 148 static_assert(BTRFS_INHIBITED_EBS_SLOTS <= 32); 149 static_assert(BTRFS_INHIBITED_EBS_SLOTS != 0 && 150 (BTRFS_INHIBITED_EBS_SLOTS & (BTRFS_INHIBITED_EBS_SLOTS - 1)) == 0); 151 152 struct btrfs_trans_handle { 153 u64 transid; 154 u64 bytes_reserved; 155 u64 delayed_refs_bytes_reserved; 156 u64 chunk_bytes_reserved; 157 unsigned long delayed_ref_updates; 158 unsigned long delayed_ref_csum_deletions; 159 struct btrfs_transaction *transaction; 160 struct btrfs_block_rsv *block_rsv; 161 struct btrfs_block_rsv *orig_rsv; 162 /* Set by a task that wants to create a snapshot. */ 163 struct btrfs_pending_snapshot *pending_snapshot; 164 refcount_t use_count; 165 unsigned int type; 166 /* 167 * Error code of transaction abort, set outside of locks and must use 168 * the READ_ONCE/WRITE_ONCE access 169 */ 170 short aborted; 171 bool adding_csums; 172 bool allocating_chunk; 173 bool removing_chunk; 174 bool reloc_reserved; 175 bool in_fsync; 176 struct btrfs_fs_info *fs_info; 177 struct list_head new_bgs; 178 struct btrfs_block_rsv delayed_rsv; 179 180 /* Extent buffers this handle has inhibited writeback on. */ 181 struct extent_buffer *inhibited_ebs[BTRFS_INHIBITED_EBS_SLOTS]; 182 /* CLOCK reference bit per slot. */ 183 u32 inhibited_ebs_referenced; 184 u32 nr_inhibited_ebs; 185 /* CLOCK hand. */ 186 u32 inhibited_ebs_hand; 187 }; 188 189 /* 190 * The abort status can be changed between calls and is not protected by locks. 191 * This accepts btrfs_transaction and btrfs_trans_handle as types. Once it's 192 * set to a non-zero value it does not change, so the macro should be in checks 193 * but is not necessary for further reads of the value. 194 */ 195 #define TRANS_ABORTED(trans) (unlikely(READ_ONCE((trans)->aborted))) 196 197 struct btrfs_pending_snapshot { 198 struct dentry *dentry; 199 struct btrfs_inode *dir; 200 struct btrfs_root *root; 201 struct btrfs_root_item *root_item; 202 struct btrfs_root *snap; 203 struct btrfs_qgroup_inherit *inherit; 204 struct btrfs_path *path; 205 /* block reservation for the operation */ 206 struct btrfs_block_rsv block_rsv; 207 /* extra metadata reservation for relocation */ 208 int error; 209 /* Preallocated anonymous block device number */ 210 dev_t anon_dev; 211 bool readonly; 212 struct list_head list; 213 }; 214 215 static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans, 216 struct btrfs_inode *inode) 217 { 218 spin_lock(&inode->lock); 219 inode->last_trans = trans->transaction->transid; 220 inode->last_sub_trans = btrfs_get_root_log_transid(inode->root); 221 inode->last_log_commit = inode->last_sub_trans - 1; 222 spin_unlock(&inode->lock); 223 } 224 225 /* 226 * Make qgroup codes to skip given qgroupid, means the old/new_roots for 227 * qgroup won't contain the qgroupid in it. 228 */ 229 static inline void btrfs_set_skip_qgroup(struct btrfs_trans_handle *trans, 230 u64 qgroupid) 231 { 232 struct btrfs_delayed_ref_root *delayed_refs; 233 234 delayed_refs = &trans->transaction->delayed_refs; 235 WARN_ON(delayed_refs->qgroup_to_skip); 236 delayed_refs->qgroup_to_skip = qgroupid; 237 } 238 239 static inline void btrfs_clear_skip_qgroup(struct btrfs_trans_handle *trans) 240 { 241 struct btrfs_delayed_ref_root *delayed_refs; 242 243 delayed_refs = &trans->transaction->delayed_refs; 244 WARN_ON(!delayed_refs->qgroup_to_skip); 245 delayed_refs->qgroup_to_skip = 0; 246 } 247 248 /* 249 * We want the transaction abort to print stack trace only for errors where the 250 * cause could be a bug, eg. due to ENOSPC, and not for common errors that are 251 * caused by external factors. 252 */ 253 static inline bool btrfs_abort_should_print_stack(int error) 254 { 255 switch (error) { 256 case -EIO: 257 case -EROFS: 258 case -ENOMEM: 259 return false; 260 } 261 return true; 262 } 263 264 /* 265 * Compile-time and run-time verification of error passed to transaction abort. 266 * Direct constants will be caught at compile time, errors read from variables 267 * can be caught only at run-time and will warn under debugging config. 268 * 269 * How verification works: 270 * - accepted builtin constants are all -EIO and such 271 * - for compile-time check, invalid condition produces a negative-sized array 272 * type, valid zero-sized 273 * - when a variable is passed as error the first check is a no-op 274 * - with enabled debugging, the second array type size is constructed from the 275 * real variable value, valid condition produces array of size 1 276 * - sizeof(type) does not generate any code 277 */ 278 #define VERIFY_NEGATIVE_ERROR(error) \ 279 do { \ 280 (void)sizeof(char[-!(__builtin_constant_p(error) ? (error) < 0 : 1)]); \ 281 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) { \ 282 if (sizeof(char[(error) < 0]) != 1) \ 283 DEBUG_WARN("error >= 0 passed to btrfs_abort_transaction()"); \ 284 } \ 285 } while(0) 286 287 /* 288 * Call btrfs_abort_transaction() as early as possible when an error condition 289 * is detected, that way the exact stack trace is reported for some errors. 290 * 291 * Error number must be negative as it encodes wheather it's the first abort. 292 */ 293 #define btrfs_abort_transaction(trans, error) \ 294 do { \ 295 int __error = (error); \ 296 \ 297 VERIFY_NEGATIVE_ERROR(error); \ 298 /* Report first abort since mount */ \ 299 if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED, \ 300 &((trans)->fs_info->fs_state))) { \ 301 WARN_ON(btrfs_abort_should_print_stack(__error)); \ 302 __error = -__error; \ 303 } \ 304 __btrfs_abort_transaction((trans), __func__, \ 305 __LINE__, __error); \ 306 } while (0) 307 308 int btrfs_end_transaction(struct btrfs_trans_handle *trans); 309 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 310 unsigned int num_items); 311 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 312 struct btrfs_root *root, 313 unsigned int num_items); 314 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root); 315 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root); 316 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root); 317 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root); 318 struct btrfs_trans_handle *btrfs_attach_transaction_barrier( 319 struct btrfs_root *root); 320 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid); 321 322 void btrfs_add_dead_root(struct btrfs_root *root); 323 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info); 324 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info); 325 int btrfs_commit_transaction(struct btrfs_trans_handle *trans); 326 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans); 327 int btrfs_commit_current_transaction(struct btrfs_root *root); 328 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans); 329 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans); 330 void btrfs_throttle(struct btrfs_fs_info *fs_info); 331 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 332 struct btrfs_root *root); 333 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 334 struct extent_io_tree *dirty_pages, int mark); 335 int btrfs_wait_tree_log_extents(struct btrfs_root *root, int mark); 336 int btrfs_transaction_blocked(struct btrfs_fs_info *info); 337 void btrfs_put_transaction(struct btrfs_transaction *transaction); 338 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 339 struct btrfs_root *root); 340 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans); 341 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans, 342 const char *function, 343 unsigned int line, int error); 344 345 int __init btrfs_transaction_init(void); 346 void __cold btrfs_transaction_exit(void); 347 348 #endif 349