~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/xfs/xfs_log_priv.h

Version: ~ [ linux-5.13-rc5 ] ~ [ linux-5.12.9 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.42 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.124 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.193 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.235 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.271 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.271 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
  3  * All Rights Reserved.
  4  *
  5  * This program is free software; you can redistribute it and/or
  6  * modify it under the terms of the GNU General Public License as
  7  * published by the Free Software Foundation.
  8  *
  9  * This program is distributed in the hope that it would be useful,
 10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
 11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 12  * GNU General Public License for more details.
 13  *
 14  * You should have received a copy of the GNU General Public License
 15  * along with this program; if not, write the Free Software Foundation,
 16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 17  */
 18 #ifndef __XFS_LOG_PRIV_H__
 19 #define __XFS_LOG_PRIV_H__
 20 
 21 struct xfs_buf;
 22 struct xlog;
 23 struct xlog_ticket;
 24 struct xfs_mount;
 25 struct xfs_log_callback;
 26 
 27 /*
 28  * Flags for log structure
 29  */
 30 #define XLOG_ACTIVE_RECOVERY    0x2     /* in the middle of recovery */
 31 #define XLOG_RECOVERY_NEEDED    0x4     /* log was recovered */
 32 #define XLOG_IO_ERROR           0x8     /* log hit an I/O error, and being
 33                                            shutdown */
 34 #define XLOG_TAIL_WARN          0x10    /* log tail verify warning issued */
 35 
 36 /*
 37  * get client id from packed copy.
 38  *
 39  * this hack is here because the xlog_pack code copies four bytes
 40  * of xlog_op_header containing the fields oh_clientid, oh_flags
 41  * and oh_res2 into the packed copy.
 42  *
 43  * later on this four byte chunk is treated as an int and the
 44  * client id is pulled out.
 45  *
 46  * this has endian issues, of course.
 47  */
 48 static inline uint xlog_get_client_id(__be32 i)
 49 {
 50         return be32_to_cpu(i) >> 24;
 51 }
 52 
 53 /*
 54  * In core log state
 55  */
 56 #define XLOG_STATE_ACTIVE    0x0001 /* Current IC log being written to */
 57 #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
 58 #define XLOG_STATE_SYNCING   0x0004 /* This IC log is syncing */
 59 #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
 60 #define XLOG_STATE_DO_CALLBACK \
 61                              0x0010 /* Process callback functions */
 62 #define XLOG_STATE_CALLBACK  0x0020 /* Callback functions now */
 63 #define XLOG_STATE_DIRTY     0x0040 /* Dirty IC log, not ready for ACTIVE status*/
 64 #define XLOG_STATE_IOERROR   0x0080 /* IO error happened in sync'ing log */
 65 #define XLOG_STATE_IOABORT   0x0100 /* force abort on I/O completion (debug) */
 66 #define XLOG_STATE_ALL       0x7FFF /* All possible valid flags */
 67 #define XLOG_STATE_NOTUSED   0x8000 /* This IC log not being used */
 68 
 69 /*
 70  * Flags to log ticket
 71  */
 72 #define XLOG_TIC_INITED         0x1     /* has been initialized */
 73 #define XLOG_TIC_PERM_RESERV    0x2     /* permanent reservation */
 74 
 75 #define XLOG_TIC_FLAGS \
 76         { XLOG_TIC_INITED,      "XLOG_TIC_INITED" }, \
 77         { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
 78 
 79 /*
 80  * Below are states for covering allocation transactions.
 81  * By covering, we mean changing the h_tail_lsn in the last on-disk
 82  * log write such that no allocation transactions will be re-done during
 83  * recovery after a system crash. Recovery starts at the last on-disk
 84  * log write.
 85  *
 86  * These states are used to insert dummy log entries to cover
 87  * space allocation transactions which can undo non-transactional changes
 88  * after a crash. Writes to a file with space
 89  * already allocated do not result in any transactions. Allocations
 90  * might include space beyond the EOF. So if we just push the EOF a
 91  * little, the last transaction for the file could contain the wrong
 92  * size. If there is no file system activity, after an allocation
 93  * transaction, and the system crashes, the allocation transaction
 94  * will get replayed and the file will be truncated. This could
 95  * be hours/days/... after the allocation occurred.
 96  *
 97  * The fix for this is to do two dummy transactions when the
 98  * system is idle. We need two dummy transaction because the h_tail_lsn
 99  * in the log record header needs to point beyond the last possible
100  * non-dummy transaction. The first dummy changes the h_tail_lsn to
101  * the first transaction before the dummy. The second dummy causes
102  * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
103  *
104  * These dummy transactions get committed when everything
105  * is idle (after there has been some activity).
106  *
107  * There are 5 states used to control this.
108  *
109  *  IDLE -- no logging has been done on the file system or
110  *              we are done covering previous transactions.
111  *  NEED -- logging has occurred and we need a dummy transaction
112  *              when the log becomes idle.
113  *  DONE -- we were in the NEED state and have committed a dummy
114  *              transaction.
115  *  NEED2 -- we detected that a dummy transaction has gone to the
116  *              on disk log with no other transactions.
117  *  DONE2 -- we committed a dummy transaction when in the NEED2 state.
118  *
119  * There are two places where we switch states:
120  *
121  * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
122  *      We commit the dummy transaction and switch to DONE or DONE2,
123  *      respectively. In all other states, we don't do anything.
124  *
125  * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
126  *
127  *      No matter what state we are in, if this isn't the dummy
128  *      transaction going out, the next state is NEED.
129  *      So, if we aren't in the DONE or DONE2 states, the next state
130  *      is NEED. We can't be finishing a write of the dummy record
131  *      unless it was committed and the state switched to DONE or DONE2.
132  *
133  *      If we are in the DONE state and this was a write of the
134  *              dummy transaction, we move to NEED2.
135  *
136  *      If we are in the DONE2 state and this was a write of the
137  *              dummy transaction, we move to IDLE.
138  *
139  *
140  * Writing only one dummy transaction can get appended to
141  * one file space allocation. When this happens, the log recovery
142  * code replays the space allocation and a file could be truncated.
143  * This is why we have the NEED2 and DONE2 states before going idle.
144  */
145 
146 #define XLOG_STATE_COVER_IDLE   0
147 #define XLOG_STATE_COVER_NEED   1
148 #define XLOG_STATE_COVER_DONE   2
149 #define XLOG_STATE_COVER_NEED2  3
150 #define XLOG_STATE_COVER_DONE2  4
151 
152 #define XLOG_COVER_OPS          5
153 
154 /* Ticket reservation region accounting */ 
155 #define XLOG_TIC_LEN_MAX        15
156 
157 /*
158  * Reservation region
159  * As would be stored in xfs_log_iovec but without the i_addr which
160  * we don't care about.
161  */
162 typedef struct xlog_res {
163         uint    r_len;  /* region length                :4 */
164         uint    r_type; /* region's transaction type    :4 */
165 } xlog_res_t;
166 
167 typedef struct xlog_ticket {
168         struct list_head   t_queue;      /* reserve/write queue */
169         struct task_struct *t_task;      /* task that owns this ticket */
170         xlog_tid_t         t_tid;        /* transaction identifier       : 4  */
171         atomic_t           t_ref;        /* ticket reference count       : 4  */
172         int                t_curr_res;   /* current reservation in bytes : 4  */
173         int                t_unit_res;   /* unit reservation in bytes    : 4  */
174         char               t_ocnt;       /* original count               : 1  */
175         char               t_cnt;        /* current count                : 1  */
176         char               t_clientid;   /* who does this belong to;     : 1  */
177         char               t_flags;      /* properties of reservation    : 1  */
178 
179         /* reservation array fields */
180         uint               t_res_num;                    /* num in array : 4 */
181         uint               t_res_num_ophdrs;             /* num op hdrs  : 4 */
182         uint               t_res_arr_sum;                /* array sum    : 4 */
183         uint               t_res_o_flow;                 /* sum overflow : 4 */
184         xlog_res_t         t_res_arr[XLOG_TIC_LEN_MAX];  /* array of res : 8 * 15 */ 
185 } xlog_ticket_t;
186 
187 /*
188  * - A log record header is 512 bytes.  There is plenty of room to grow the
189  *      xlog_rec_header_t into the reserved space.
190  * - ic_data follows, so a write to disk can start at the beginning of
191  *      the iclog.
192  * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
193  * - ic_next is the pointer to the next iclog in the ring.
194  * - ic_bp is a pointer to the buffer used to write this incore log to disk.
195  * - ic_log is a pointer back to the global log structure.
196  * - ic_callback is a linked list of callback function/argument pairs to be
197  *      called after an iclog finishes writing.
198  * - ic_size is the full size of the header plus data.
199  * - ic_offset is the current number of bytes written to in this iclog.
200  * - ic_refcnt is bumped when someone is writing to the log.
201  * - ic_state is the state of the iclog.
202  *
203  * Because of cacheline contention on large machines, we need to separate
204  * various resources onto different cachelines. To start with, make the
205  * structure cacheline aligned. The following fields can be contended on
206  * by independent processes:
207  *
208  *      - ic_callback_*
209  *      - ic_refcnt
210  *      - fields protected by the global l_icloglock
211  *
212  * so we need to ensure that these fields are located in separate cachelines.
213  * We'll put all the read-only and l_icloglock fields in the first cacheline,
214  * and move everything else out to subsequent cachelines.
215  */
216 typedef struct xlog_in_core {
217         wait_queue_head_t       ic_force_wait;
218         wait_queue_head_t       ic_write_wait;
219         struct xlog_in_core     *ic_next;
220         struct xlog_in_core     *ic_prev;
221         struct xfs_buf          *ic_bp;
222         struct xlog             *ic_log;
223         int                     ic_size;
224         int                     ic_offset;
225         int                     ic_bwritecnt;
226         unsigned short          ic_state;
227         char                    *ic_datap;      /* pointer to iclog data */
228 
229         /* Callback structures need their own cacheline */
230         spinlock_t              ic_callback_lock ____cacheline_aligned_in_smp;
231         struct xfs_log_callback *ic_callback;
232         struct xfs_log_callback **ic_callback_tail;
233 
234         /* reference counts need their own cacheline */
235         atomic_t                ic_refcnt ____cacheline_aligned_in_smp;
236         xlog_in_core_2_t        *ic_data;
237 #define ic_header       ic_data->hic_header
238 } xlog_in_core_t;
239 
240 /*
241  * The CIL context is used to aggregate per-transaction details as well be
242  * passed to the iclog for checkpoint post-commit processing.  After being
243  * passed to the iclog, another context needs to be allocated for tracking the
244  * next set of transactions to be aggregated into a checkpoint.
245  */
246 struct xfs_cil;
247 
248 struct xfs_cil_ctx {
249         struct xfs_cil          *cil;
250         xfs_lsn_t               sequence;       /* chkpt sequence # */
251         xfs_lsn_t               start_lsn;      /* first LSN of chkpt commit */
252         xfs_lsn_t               commit_lsn;     /* chkpt commit record lsn */
253         struct xlog_ticket      *ticket;        /* chkpt ticket */
254         int                     nvecs;          /* number of regions */
255         int                     space_used;     /* aggregate size of regions */
256         struct list_head        busy_extents;   /* busy extents in chkpt */
257         struct xfs_log_vec      *lv_chain;      /* logvecs being pushed */
258         struct xfs_log_callback log_cb;         /* completion callback hook. */
259         struct list_head        committing;     /* ctx committing list */
260         struct work_struct      discard_endio_work;
261 };
262 
263 /*
264  * Committed Item List structure
265  *
266  * This structure is used to track log items that have been committed but not
267  * yet written into the log. It is used only when the delayed logging mount
268  * option is enabled.
269  *
270  * This structure tracks the list of committing checkpoint contexts so
271  * we can avoid the problem of having to hold out new transactions during a
272  * flush until we have a the commit record LSN of the checkpoint. We can
273  * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
274  * sequence match and extract the commit LSN directly from there. If the
275  * checkpoint is still in the process of committing, we can block waiting for
276  * the commit LSN to be determined as well. This should make synchronous
277  * operations almost as efficient as the old logging methods.
278  */
279 struct xfs_cil {
280         struct xlog             *xc_log;
281         struct list_head        xc_cil;
282         spinlock_t              xc_cil_lock;
283 
284         struct rw_semaphore     xc_ctx_lock ____cacheline_aligned_in_smp;
285         struct xfs_cil_ctx      *xc_ctx;
286 
287         spinlock_t              xc_push_lock ____cacheline_aligned_in_smp;
288         xfs_lsn_t               xc_push_seq;
289         struct list_head        xc_committing;
290         wait_queue_head_t       xc_commit_wait;
291         xfs_lsn_t               xc_current_sequence;
292         struct work_struct      xc_push_work;
293 } ____cacheline_aligned_in_smp;
294 
295 /*
296  * The amount of log space we allow the CIL to aggregate is difficult to size.
297  * Whatever we choose, we have to make sure we can get a reservation for the
298  * log space effectively, that it is large enough to capture sufficient
299  * relogging to reduce log buffer IO significantly, but it is not too large for
300  * the log or induces too much latency when writing out through the iclogs. We
301  * track both space consumed and the number of vectors in the checkpoint
302  * context, so we need to decide which to use for limiting.
303  *
304  * Every log buffer we write out during a push needs a header reserved, which
305  * is at least one sector and more for v2 logs. Hence we need a reservation of
306  * at least 512 bytes per 32k of log space just for the LR headers. That means
307  * 16KB of reservation per megabyte of delayed logging space we will consume,
308  * plus various headers.  The number of headers will vary based on the num of
309  * io vectors, so limiting on a specific number of vectors is going to result
310  * in transactions of varying size. IOWs, it is more consistent to track and
311  * limit space consumed in the log rather than by the number of objects being
312  * logged in order to prevent checkpoint ticket overruns.
313  *
314  * Further, use of static reservations through the log grant mechanism is
315  * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
316  * grant) and a significant deadlock potential because regranting write space
317  * can block on log pushes. Hence if we have to regrant log space during a log
318  * push, we can deadlock.
319  *
320  * However, we can avoid this by use of a dynamic "reservation stealing"
321  * technique during transaction commit whereby unused reservation space in the
322  * transaction ticket is transferred to the CIL ctx commit ticket to cover the
323  * space needed by the checkpoint transaction. This means that we never need to
324  * specifically reserve space for the CIL checkpoint transaction, nor do we
325  * need to regrant space once the checkpoint completes. This also means the
326  * checkpoint transaction ticket is specific to the checkpoint context, rather
327  * than the CIL itself.
328  *
329  * With dynamic reservations, we can effectively make up arbitrary limits for
330  * the checkpoint size so long as they don't violate any other size rules.
331  * Recovery imposes a rule that no transaction exceed half the log, so we are
332  * limited by that.  Furthermore, the log transaction reservation subsystem
333  * tries to keep 25% of the log free, so we need to keep below that limit or we
334  * risk running out of free log space to start any new transactions.
335  *
336  * In order to keep background CIL push efficient, we will set a lower
337  * threshold at which background pushing is attempted without blocking current
338  * transaction commits.  A separate, higher bound defines when CIL pushes are
339  * enforced to ensure we stay within our maximum checkpoint size bounds.
340  * threshold, yet give us plenty of space for aggregation on large logs.
341  */
342 #define XLOG_CIL_SPACE_LIMIT(log)       (log->l_logsize >> 3)
343 
344 /*
345  * ticket grant locks, queues and accounting have their own cachlines
346  * as these are quite hot and can be operated on concurrently.
347  */
348 struct xlog_grant_head {
349         spinlock_t              lock ____cacheline_aligned_in_smp;
350         struct list_head        waiters;
351         atomic64_t              grant;
352 };
353 
354 /*
355  * The reservation head lsn is not made up of a cycle number and block number.
356  * Instead, it uses a cycle number and byte number.  Logs don't expect to
357  * overflow 31 bits worth of byte offset, so using a byte number will mean
358  * that round off problems won't occur when releasing partial reservations.
359  */
360 struct xlog {
361         /* The following fields don't need locking */
362         struct xfs_mount        *l_mp;          /* mount point */
363         struct xfs_ail          *l_ailp;        /* AIL log is working with */
364         struct xfs_cil          *l_cilp;        /* CIL log is working with */
365         struct xfs_buf          *l_xbuf;        /* extra buffer for log
366                                                  * wrapping */
367         struct xfs_buftarg      *l_targ;        /* buftarg of log */
368         struct delayed_work     l_work;         /* background flush work */
369         uint                    l_flags;
370         uint                    l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
371         struct list_head        *l_buf_cancel_table;
372         int                     l_iclog_hsize;  /* size of iclog header */
373         int                     l_iclog_heads;  /* # of iclog header sectors */
374         uint                    l_sectBBsize;   /* sector size in BBs (2^n) */
375         int                     l_iclog_size;   /* size of log in bytes */
376         int                     l_iclog_size_log; /* log power size of log */
377         int                     l_iclog_bufs;   /* number of iclog buffers */
378         xfs_daddr_t             l_logBBstart;   /* start block of log */
379         int                     l_logsize;      /* size of log in bytes */
380         int                     l_logBBsize;    /* size of log in BB chunks */
381 
382         /* The following block of fields are changed while holding icloglock */
383         wait_queue_head_t       l_flush_wait ____cacheline_aligned_in_smp;
384                                                 /* waiting for iclog flush */
385         int                     l_covered_state;/* state of "covering disk
386                                                  * log entries" */
387         xlog_in_core_t          *l_iclog;       /* head log queue       */
388         spinlock_t              l_icloglock;    /* grab to change iclog state */
389         int                     l_curr_cycle;   /* Cycle number of log writes */
390         int                     l_prev_cycle;   /* Cycle number before last
391                                                  * block increment */
392         int                     l_curr_block;   /* current logical log block */
393         int                     l_prev_block;   /* previous logical log block */
394 
395         /*
396          * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
397          * read without needing to hold specific locks. To avoid operations
398          * contending with other hot objects, place each of them on a separate
399          * cacheline.
400          */
401         /* lsn of last LR on disk */
402         atomic64_t              l_last_sync_lsn ____cacheline_aligned_in_smp;
403         /* lsn of 1st LR with unflushed * buffers */
404         atomic64_t              l_tail_lsn ____cacheline_aligned_in_smp;
405 
406         struct xlog_grant_head  l_reserve_head;
407         struct xlog_grant_head  l_write_head;
408 
409         struct xfs_kobj         l_kobj;
410 
411         /* The following field are used for debugging; need to hold icloglock */
412 #ifdef DEBUG
413         void                    *l_iclog_bak[XLOG_MAX_ICLOGS];
414         /* log record crc error injection factor */
415         uint32_t                l_badcrc_factor;
416 #endif
417         /* log recovery lsn tracking (for buffer submission */
418         xfs_lsn_t               l_recovery_lsn;
419 };
420 
421 #define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
422         ((log)->l_buf_cancel_table + ((uint64_t)blkno % XLOG_BC_TABLE_SIZE))
423 
424 #define XLOG_FORCED_SHUTDOWN(log)       ((log)->l_flags & XLOG_IO_ERROR)
425 
426 /* common routines */
427 extern int
428 xlog_recover(
429         struct xlog             *log);
430 extern int
431 xlog_recover_finish(
432         struct xlog             *log);
433 extern int
434 xlog_recover_cancel(struct xlog *);
435 
436 extern __le32    xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
437                             char *dp, int size);
438 
439 extern kmem_zone_t *xfs_log_ticket_zone;
440 struct xlog_ticket *
441 xlog_ticket_alloc(
442         struct xlog     *log,
443         int             unit_bytes,
444         int             count,
445         char            client,
446         bool            permanent,
447         xfs_km_flags_t  alloc_flags);
448 
449 
450 static inline void
451 xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
452 {
453         *ptr += bytes;
454         *len -= bytes;
455         *off += bytes;
456 }
457 
458 void    xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
459 void    xlog_print_trans(struct xfs_trans *);
460 int
461 xlog_write(
462         struct xlog             *log,
463         struct xfs_log_vec      *log_vector,
464         struct xlog_ticket      *tic,
465         xfs_lsn_t               *start_lsn,
466         struct xlog_in_core     **commit_iclog,
467         uint                    flags);
468 
469 /*
470  * When we crack an atomic LSN, we sample it first so that the value will not
471  * change while we are cracking it into the component values. This means we
472  * will always get consistent component values to work from. This should always
473  * be used to sample and crack LSNs that are stored and updated in atomic
474  * variables.
475  */
476 static inline void
477 xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
478 {
479         xfs_lsn_t val = atomic64_read(lsn);
480 
481         *cycle = CYCLE_LSN(val);
482         *block = BLOCK_LSN(val);
483 }
484 
485 /*
486  * Calculate and assign a value to an atomic LSN variable from component pieces.
487  */
488 static inline void
489 xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
490 {
491         atomic64_set(lsn, xlog_assign_lsn(cycle, block));
492 }
493 
494 /*
495  * When we crack the grant head, we sample it first so that the value will not
496  * change while we are cracking it into the component values. This means we
497  * will always get consistent component values to work from.
498  */
499 static inline void
500 xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
501 {
502         *cycle = val >> 32;
503         *space = val & 0xffffffff;
504 }
505 
506 static inline void
507 xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
508 {
509         xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
510 }
511 
512 static inline int64_t
513 xlog_assign_grant_head_val(int cycle, int space)
514 {
515         return ((int64_t)cycle << 32) | space;
516 }
517 
518 static inline void
519 xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
520 {
521         atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
522 }
523 
524 /*
525  * Committed Item List interfaces
526  */
527 int     xlog_cil_init(struct xlog *log);
528 void    xlog_cil_init_post_recovery(struct xlog *log);
529 void    xlog_cil_destroy(struct xlog *log);
530 bool    xlog_cil_empty(struct xlog *log);
531 
532 /*
533  * CIL force routines
534  */
535 xfs_lsn_t
536 xlog_cil_force_lsn(
537         struct xlog *log,
538         xfs_lsn_t sequence);
539 
540 static inline void
541 xlog_cil_force(struct xlog *log)
542 {
543         xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
544 }
545 
546 /*
547  * Unmount record type is used as a pseudo transaction type for the ticket.
548  * It's value must be outside the range of XFS_TRANS_* values.
549  */
550 #define XLOG_UNMOUNT_REC_TYPE   (-1U)
551 
552 /*
553  * Wrapper function for waiting on a wait queue serialised against wakeups
554  * by a spinlock. This matches the semantics of all the wait queues used in the
555  * log code.
556  */
557 static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock)
558 {
559         DECLARE_WAITQUEUE(wait, current);
560 
561         add_wait_queue_exclusive(wq, &wait);
562         __set_current_state(TASK_UNINTERRUPTIBLE);
563         spin_unlock(lock);
564         schedule();
565         remove_wait_queue(wq, &wait);
566 }
567 
568 /*
569  * The LSN is valid so long as it is behind the current LSN. If it isn't, this
570  * means that the next log record that includes this metadata could have a
571  * smaller LSN. In turn, this means that the modification in the log would not
572  * replay.
573  */
574 static inline bool
575 xlog_valid_lsn(
576         struct xlog     *log,
577         xfs_lsn_t       lsn)
578 {
579         int             cur_cycle;
580         int             cur_block;
581         bool            valid = true;
582 
583         /*
584          * First, sample the current lsn without locking to avoid added
585          * contention from metadata I/O. The current cycle and block are updated
586          * (in xlog_state_switch_iclogs()) and read here in a particular order
587          * to avoid false negatives (e.g., thinking the metadata LSN is valid
588          * when it is not).
589          *
590          * The current block is always rewound before the cycle is bumped in
591          * xlog_state_switch_iclogs() to ensure the current LSN is never seen in
592          * a transiently forward state. Instead, we can see the LSN in a
593          * transiently behind state if we happen to race with a cycle wrap.
594          */
595         cur_cycle = READ_ONCE(log->l_curr_cycle);
596         smp_rmb();
597         cur_block = READ_ONCE(log->l_curr_block);
598 
599         if ((CYCLE_LSN(lsn) > cur_cycle) ||
600             (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) {
601                 /*
602                  * If the metadata LSN appears invalid, it's possible the check
603                  * above raced with a wrap to the next log cycle. Grab the lock
604                  * to check for sure.
605                  */
606                 spin_lock(&log->l_icloglock);
607                 cur_cycle = log->l_curr_cycle;
608                 cur_block = log->l_curr_block;
609                 spin_unlock(&log->l_icloglock);
610 
611                 if ((CYCLE_LSN(lsn) > cur_cycle) ||
612                     (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block))
613                         valid = false;
614         }
615 
616         return valid;
617 }
618 
619 #endif  /* __XFS_LOG_PRIV_H__ */
620 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp