Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
0004  * Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
0005  */
0006 
0007 #include <linux/sched.h>
0008 #include <linux/slab.h>
0009 #include <linux/spinlock.h>
0010 #include <linux/completion.h>
0011 #include <linux/buffer_head.h>
0012 #include <linux/gfs2_ondisk.h>
0013 #include <linux/crc32.h>
0014 #include <linux/crc32c.h>
0015 #include <linux/delay.h>
0016 #include <linux/kthread.h>
0017 #include <linux/freezer.h>
0018 #include <linux/bio.h>
0019 #include <linux/blkdev.h>
0020 #include <linux/writeback.h>
0021 #include <linux/list_sort.h>
0022 
0023 #include "gfs2.h"
0024 #include "incore.h"
0025 #include "bmap.h"
0026 #include "glock.h"
0027 #include "log.h"
0028 #include "lops.h"
0029 #include "meta_io.h"
0030 #include "util.h"
0031 #include "dir.h"
0032 #include "trace_gfs2.h"
0033 #include "trans.h"
0034 
0035 static void gfs2_log_shutdown(struct gfs2_sbd *sdp);
0036 
0037 /**
0038  * gfs2_struct2blk - compute stuff
0039  * @sdp: the filesystem
0040  * @nstruct: the number of structures
0041  *
0042  * Compute the number of log descriptor blocks needed to hold a certain number
0043  * of structures of a certain size.
0044  *
0045  * Returns: the number of blocks needed (minimum is always 1)
0046  */
0047 
0048 unsigned int gfs2_struct2blk(struct gfs2_sbd *sdp, unsigned int nstruct)
0049 {
0050     unsigned int blks;
0051     unsigned int first, second;
0052 
0053     /* The initial struct gfs2_log_descriptor block */
0054     blks = 1;
0055     first = sdp->sd_ldptrs;
0056 
0057     if (nstruct > first) {
0058         /* Subsequent struct gfs2_meta_header blocks */
0059         second = sdp->sd_inptrs;
0060         blks += DIV_ROUND_UP(nstruct - first, second);
0061     }
0062 
0063     return blks;
0064 }
0065 
0066 /**
0067  * gfs2_remove_from_ail - Remove an entry from the ail lists, updating counters
0068  * @bd: The gfs2_bufdata to remove
0069  *
0070  * The ail lock _must_ be held when calling this function
0071  *
0072  */
0073 
0074 void gfs2_remove_from_ail(struct gfs2_bufdata *bd)
0075 {
0076     bd->bd_tr = NULL;
0077     list_del_init(&bd->bd_ail_st_list);
0078     list_del_init(&bd->bd_ail_gl_list);
0079     atomic_dec(&bd->bd_gl->gl_ail_count);
0080     brelse(bd->bd_bh);
0081 }
0082 
0083 /**
0084  * gfs2_ail1_start_one - Start I/O on a transaction
0085  * @sdp: The superblock
0086  * @wbc: The writeback control structure
0087  * @tr: The transaction to start I/O on
0088  * @plug: The block plug currently active
0089  */
0090 
0091 static int gfs2_ail1_start_one(struct gfs2_sbd *sdp,
0092                    struct writeback_control *wbc,
0093                    struct gfs2_trans *tr, struct blk_plug *plug)
0094 __releases(&sdp->sd_ail_lock)
0095 __acquires(&sdp->sd_ail_lock)
0096 {
0097     struct gfs2_glock *gl = NULL;
0098     struct address_space *mapping;
0099     struct gfs2_bufdata *bd, *s;
0100     struct buffer_head *bh;
0101     int ret = 0;
0102 
0103     list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list, bd_ail_st_list) {
0104         bh = bd->bd_bh;
0105 
0106         gfs2_assert(sdp, bd->bd_tr == tr);
0107 
0108         if (!buffer_busy(bh)) {
0109             if (buffer_uptodate(bh)) {
0110                 list_move(&bd->bd_ail_st_list,
0111                       &tr->tr_ail2_list);
0112                 continue;
0113             }
0114             if (!cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
0115                 gfs2_io_error_bh(sdp, bh);
0116                 gfs2_withdraw_delayed(sdp);
0117             }
0118         }
0119 
0120         if (gfs2_withdrawn(sdp)) {
0121             gfs2_remove_from_ail(bd);
0122             continue;
0123         }
0124         if (!buffer_dirty(bh))
0125             continue;
0126         if (gl == bd->bd_gl)
0127             continue;
0128         gl = bd->bd_gl;
0129         list_move(&bd->bd_ail_st_list, &tr->tr_ail1_list);
0130         mapping = bh->b_page->mapping;
0131         if (!mapping)
0132             continue;
0133         spin_unlock(&sdp->sd_ail_lock);
0134         ret = filemap_fdatawrite_wbc(mapping, wbc);
0135         if (need_resched()) {
0136             blk_finish_plug(plug);
0137             cond_resched();
0138             blk_start_plug(plug);
0139         }
0140         spin_lock(&sdp->sd_ail_lock);
0141         if (ret == -ENODATA) /* if a jdata write into a new hole */
0142             ret = 0; /* ignore it */
0143         if (ret || wbc->nr_to_write <= 0)
0144             break;
0145         return -EBUSY;
0146     }
0147 
0148     return ret;
0149 }
0150 
0151 static void dump_ail_list(struct gfs2_sbd *sdp)
0152 {
0153     struct gfs2_trans *tr;
0154     struct gfs2_bufdata *bd;
0155     struct buffer_head *bh;
0156 
0157     list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
0158         list_for_each_entry_reverse(bd, &tr->tr_ail1_list,
0159                         bd_ail_st_list) {
0160             bh = bd->bd_bh;
0161             fs_err(sdp, "bd %p: blk:0x%llx bh=%p ", bd,
0162                    (unsigned long long)bd->bd_blkno, bh);
0163             if (!bh) {
0164                 fs_err(sdp, "\n");
0165                 continue;
0166             }
0167             fs_err(sdp, "0x%llx up2:%d dirt:%d lkd:%d req:%d "
0168                    "map:%d new:%d ar:%d aw:%d delay:%d "
0169                    "io err:%d unwritten:%d dfr:%d pin:%d esc:%d\n",
0170                    (unsigned long long)bh->b_blocknr,
0171                    buffer_uptodate(bh), buffer_dirty(bh),
0172                    buffer_locked(bh), buffer_req(bh),
0173                    buffer_mapped(bh), buffer_new(bh),
0174                    buffer_async_read(bh), buffer_async_write(bh),
0175                    buffer_delay(bh), buffer_write_io_error(bh),
0176                    buffer_unwritten(bh),
0177                    buffer_defer_completion(bh),
0178                    buffer_pinned(bh), buffer_escaped(bh));
0179         }
0180     }
0181 }
0182 
0183 /**
0184  * gfs2_ail1_flush - start writeback of some ail1 entries 
0185  * @sdp: The super block
0186  * @wbc: The writeback control structure
0187  *
0188  * Writes back some ail1 entries, according to the limits in the
0189  * writeback control structure
0190  */
0191 
0192 void gfs2_ail1_flush(struct gfs2_sbd *sdp, struct writeback_control *wbc)
0193 {
0194     struct list_head *head = &sdp->sd_ail1_list;
0195     struct gfs2_trans *tr;
0196     struct blk_plug plug;
0197     int ret;
0198     unsigned long flush_start = jiffies;
0199 
0200     trace_gfs2_ail_flush(sdp, wbc, 1);
0201     blk_start_plug(&plug);
0202     spin_lock(&sdp->sd_ail_lock);
0203 restart:
0204     ret = 0;
0205     if (time_after(jiffies, flush_start + (HZ * 600))) {
0206         fs_err(sdp, "Error: In %s for ten minutes! t=%d\n",
0207                __func__, current->journal_info ? 1 : 0);
0208         dump_ail_list(sdp);
0209         goto out;
0210     }
0211     list_for_each_entry_reverse(tr, head, tr_list) {
0212         if (wbc->nr_to_write <= 0)
0213             break;
0214         ret = gfs2_ail1_start_one(sdp, wbc, tr, &plug);
0215         if (ret) {
0216             if (ret == -EBUSY)
0217                 goto restart;
0218             break;
0219         }
0220     }
0221 out:
0222     spin_unlock(&sdp->sd_ail_lock);
0223     blk_finish_plug(&plug);
0224     if (ret) {
0225         gfs2_lm(sdp, "gfs2_ail1_start_one returned: %d\n", ret);
0226         gfs2_withdraw(sdp);
0227     }
0228     trace_gfs2_ail_flush(sdp, wbc, 0);
0229 }
0230 
0231 /**
0232  * gfs2_ail1_start - start writeback of all ail1 entries
0233  * @sdp: The superblock
0234  */
0235 
0236 static void gfs2_ail1_start(struct gfs2_sbd *sdp)
0237 {
0238     struct writeback_control wbc = {
0239         .sync_mode = WB_SYNC_NONE,
0240         .nr_to_write = LONG_MAX,
0241         .range_start = 0,
0242         .range_end = LLONG_MAX,
0243     };
0244 
0245     return gfs2_ail1_flush(sdp, &wbc);
0246 }
0247 
0248 static void gfs2_log_update_flush_tail(struct gfs2_sbd *sdp)
0249 {
0250     unsigned int new_flush_tail = sdp->sd_log_head;
0251     struct gfs2_trans *tr;
0252 
0253     if (!list_empty(&sdp->sd_ail1_list)) {
0254         tr = list_last_entry(&sdp->sd_ail1_list,
0255                      struct gfs2_trans, tr_list);
0256         new_flush_tail = tr->tr_first;
0257     }
0258     sdp->sd_log_flush_tail = new_flush_tail;
0259 }
0260 
0261 static void gfs2_log_update_head(struct gfs2_sbd *sdp)
0262 {
0263     unsigned int new_head = sdp->sd_log_flush_head;
0264 
0265     if (sdp->sd_log_flush_tail == sdp->sd_log_head)
0266         sdp->sd_log_flush_tail = new_head;
0267     sdp->sd_log_head = new_head;
0268 }
0269 
0270 /*
0271  * gfs2_ail_empty_tr - empty one of the ail lists of a transaction
0272  */
0273 
0274 static void gfs2_ail_empty_tr(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
0275                   struct list_head *head)
0276 {
0277     struct gfs2_bufdata *bd;
0278 
0279     while (!list_empty(head)) {
0280         bd = list_first_entry(head, struct gfs2_bufdata,
0281                       bd_ail_st_list);
0282         gfs2_assert(sdp, bd->bd_tr == tr);
0283         gfs2_remove_from_ail(bd);
0284     }
0285 }
0286 
0287 /**
0288  * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
0289  * @sdp: the filesystem
0290  * @tr: the transaction
0291  * @max_revokes: If nonzero, issue revokes for the bd items for written buffers
0292  *
0293  * returns: the transaction's count of remaining active items
0294  */
0295 
0296 static int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
0297                 int *max_revokes)
0298 {
0299     struct gfs2_bufdata *bd, *s;
0300     struct buffer_head *bh;
0301     int active_count = 0;
0302 
0303     list_for_each_entry_safe_reverse(bd, s, &tr->tr_ail1_list,
0304                      bd_ail_st_list) {
0305         bh = bd->bd_bh;
0306         gfs2_assert(sdp, bd->bd_tr == tr);
0307         /*
0308          * If another process flagged an io error, e.g. writing to the
0309          * journal, error all other bhs and move them off the ail1 to
0310          * prevent a tight loop when unmount tries to flush ail1,
0311          * regardless of whether they're still busy. If no outside
0312          * errors were found and the buffer is busy, move to the next.
0313          * If the ail buffer is not busy and caught an error, flag it
0314          * for others.
0315          */
0316         if (!sdp->sd_log_error && buffer_busy(bh)) {
0317             active_count++;
0318             continue;
0319         }
0320         if (!buffer_uptodate(bh) &&
0321             !cmpxchg(&sdp->sd_log_error, 0, -EIO)) {
0322             gfs2_io_error_bh(sdp, bh);
0323             gfs2_withdraw_delayed(sdp);
0324         }
0325         /*
0326          * If we have space for revokes and the bd is no longer on any
0327          * buf list, we can just add a revoke for it immediately and
0328          * avoid having to put it on the ail2 list, where it would need
0329          * to be revoked later.
0330          */
0331         if (*max_revokes && list_empty(&bd->bd_list)) {
0332             gfs2_add_revoke(sdp, bd);
0333             (*max_revokes)--;
0334             continue;
0335         }
0336         list_move(&bd->bd_ail_st_list, &tr->tr_ail2_list);
0337     }
0338     return active_count;
0339 }
0340 
0341 /**
0342  * gfs2_ail1_empty - Try to empty the ail1 lists
0343  * @sdp: The superblock
0344  * @max_revokes: If non-zero, add revokes where appropriate
0345  *
0346  * Tries to empty the ail1 lists, starting with the oldest first
0347  */
0348 
0349 static int gfs2_ail1_empty(struct gfs2_sbd *sdp, int max_revokes)
0350 {
0351     struct gfs2_trans *tr, *s;
0352     int oldest_tr = 1;
0353     int ret;
0354 
0355     spin_lock(&sdp->sd_ail_lock);
0356     list_for_each_entry_safe_reverse(tr, s, &sdp->sd_ail1_list, tr_list) {
0357         if (!gfs2_ail1_empty_one(sdp, tr, &max_revokes) && oldest_tr)
0358             list_move(&tr->tr_list, &sdp->sd_ail2_list);
0359         else
0360             oldest_tr = 0;
0361     }
0362     gfs2_log_update_flush_tail(sdp);
0363     ret = list_empty(&sdp->sd_ail1_list);
0364     spin_unlock(&sdp->sd_ail_lock);
0365 
0366     if (test_bit(SDF_WITHDRAWING, &sdp->sd_flags)) {
0367         gfs2_lm(sdp, "fatal: I/O error(s)\n");
0368         gfs2_withdraw(sdp);
0369     }
0370 
0371     return ret;
0372 }
0373 
0374 static void gfs2_ail1_wait(struct gfs2_sbd *sdp)
0375 {
0376     struct gfs2_trans *tr;
0377     struct gfs2_bufdata *bd;
0378     struct buffer_head *bh;
0379 
0380     spin_lock(&sdp->sd_ail_lock);
0381     list_for_each_entry_reverse(tr, &sdp->sd_ail1_list, tr_list) {
0382         list_for_each_entry(bd, &tr->tr_ail1_list, bd_ail_st_list) {
0383             bh = bd->bd_bh;
0384             if (!buffer_locked(bh))
0385                 continue;
0386             get_bh(bh);
0387             spin_unlock(&sdp->sd_ail_lock);
0388             wait_on_buffer(bh);
0389             brelse(bh);
0390             return;
0391         }
0392     }
0393     spin_unlock(&sdp->sd_ail_lock);
0394 }
0395 
0396 static void __ail2_empty(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
0397 {
0398     gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
0399     list_del(&tr->tr_list);
0400     gfs2_assert_warn(sdp, list_empty(&tr->tr_ail1_list));
0401     gfs2_assert_warn(sdp, list_empty(&tr->tr_ail2_list));
0402     gfs2_trans_free(sdp, tr);
0403 }
0404 
0405 static void ail2_empty(struct gfs2_sbd *sdp, unsigned int new_tail)
0406 {
0407     struct list_head *ail2_list = &sdp->sd_ail2_list;
0408     unsigned int old_tail = sdp->sd_log_tail;
0409     struct gfs2_trans *tr, *safe;
0410 
0411     spin_lock(&sdp->sd_ail_lock);
0412     if (old_tail <= new_tail) {
0413         list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
0414             if (old_tail <= tr->tr_first && tr->tr_first < new_tail)
0415                 __ail2_empty(sdp, tr);
0416         }
0417     } else {
0418         list_for_each_entry_safe(tr, safe, ail2_list, tr_list) {
0419             if (old_tail <= tr->tr_first || tr->tr_first < new_tail)
0420                 __ail2_empty(sdp, tr);
0421         }
0422     }
0423     spin_unlock(&sdp->sd_ail_lock);
0424 }
0425 
0426 /**
0427  * gfs2_log_is_empty - Check if the log is empty
0428  * @sdp: The GFS2 superblock
0429  */
0430 
0431 bool gfs2_log_is_empty(struct gfs2_sbd *sdp) {
0432     return atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks;
0433 }
0434 
0435 static bool __gfs2_log_try_reserve_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
0436 {
0437     unsigned int available;
0438 
0439     available = atomic_read(&sdp->sd_log_revokes_available);
0440     while (available >= revokes) {
0441         if (atomic_try_cmpxchg(&sdp->sd_log_revokes_available,
0442                        &available, available - revokes))
0443             return true;
0444     }
0445     return false;
0446 }
0447 
0448 /**
0449  * gfs2_log_release_revokes - Release a given number of revokes
0450  * @sdp: The GFS2 superblock
0451  * @revokes: The number of revokes to release
0452  *
0453  * sdp->sd_log_flush_lock must be held.
0454  */
0455 void gfs2_log_release_revokes(struct gfs2_sbd *sdp, unsigned int revokes)
0456 {
0457     if (revokes)
0458         atomic_add(revokes, &sdp->sd_log_revokes_available);
0459 }
0460 
0461 /**
0462  * gfs2_log_release - Release a given number of log blocks
0463  * @sdp: The GFS2 superblock
0464  * @blks: The number of blocks
0465  *
0466  */
0467 
0468 void gfs2_log_release(struct gfs2_sbd *sdp, unsigned int blks)
0469 {
0470     atomic_add(blks, &sdp->sd_log_blks_free);
0471     trace_gfs2_log_blocks(sdp, blks);
0472     gfs2_assert_withdraw(sdp, atomic_read(&sdp->sd_log_blks_free) <=
0473                   sdp->sd_jdesc->jd_blocks);
0474     if (atomic_read(&sdp->sd_log_blks_needed))
0475         wake_up(&sdp->sd_log_waitq);
0476 }
0477 
0478 /**
0479  * __gfs2_log_try_reserve - Try to make a log reservation
0480  * @sdp: The GFS2 superblock
0481  * @blks: The number of blocks to reserve
0482  * @taboo_blks: The number of blocks to leave free
0483  *
0484  * Try to do the same as __gfs2_log_reserve(), but fail if no more log
0485  * space is immediately available.
0486  */
0487 static bool __gfs2_log_try_reserve(struct gfs2_sbd *sdp, unsigned int blks,
0488                    unsigned int taboo_blks)
0489 {
0490     unsigned wanted = blks + taboo_blks;
0491     unsigned int free_blocks;
0492 
0493     free_blocks = atomic_read(&sdp->sd_log_blks_free);
0494     while (free_blocks >= wanted) {
0495         if (atomic_try_cmpxchg(&sdp->sd_log_blks_free, &free_blocks,
0496                        free_blocks - blks)) {
0497             trace_gfs2_log_blocks(sdp, -blks);
0498             return true;
0499         }
0500     }
0501     return false;
0502 }
0503 
0504 /**
0505  * __gfs2_log_reserve - Make a log reservation
0506  * @sdp: The GFS2 superblock
0507  * @blks: The number of blocks to reserve
0508  * @taboo_blks: The number of blocks to leave free
0509  *
0510  * @taboo_blks is set to 0 for logd, and to GFS2_LOG_FLUSH_MIN_BLOCKS
0511  * for all other processes.  This ensures that when the log is almost full,
0512  * logd will still be able to call gfs2_log_flush one more time  without
0513  * blocking, which will advance the tail and make some more log space
0514  * available.
0515  *
0516  * We no longer flush the log here, instead we wake up logd to do that
0517  * for us. To avoid the thundering herd and to ensure that we deal fairly
0518  * with queued waiters, we use an exclusive wait. This means that when we
0519  * get woken with enough journal space to get our reservation, we need to
0520  * wake the next waiter on the list.
0521  */
0522 
0523 static void __gfs2_log_reserve(struct gfs2_sbd *sdp, unsigned int blks,
0524                    unsigned int taboo_blks)
0525 {
0526     unsigned wanted = blks + taboo_blks;
0527     unsigned int free_blocks;
0528 
0529     atomic_add(blks, &sdp->sd_log_blks_needed);
0530     for (;;) {
0531         if (current != sdp->sd_logd_process)
0532             wake_up(&sdp->sd_logd_waitq);
0533         io_wait_event(sdp->sd_log_waitq,
0534             (free_blocks = atomic_read(&sdp->sd_log_blks_free),
0535              free_blocks >= wanted));
0536         do {
0537             if (atomic_try_cmpxchg(&sdp->sd_log_blks_free,
0538                            &free_blocks,
0539                            free_blocks - blks))
0540                 goto reserved;
0541         } while (free_blocks >= wanted);
0542     }
0543 
0544 reserved:
0545     trace_gfs2_log_blocks(sdp, -blks);
0546     if (atomic_sub_return(blks, &sdp->sd_log_blks_needed))
0547         wake_up(&sdp->sd_log_waitq);
0548 }
0549 
0550 /**
0551  * gfs2_log_try_reserve - Try to make a log reservation
0552  * @sdp: The GFS2 superblock
0553  * @tr: The transaction
0554  * @extra_revokes: The number of additional revokes reserved (output)
0555  *
0556  * This is similar to gfs2_log_reserve, but sdp->sd_log_flush_lock must be
0557  * held for correct revoke accounting.
0558  */
0559 
0560 bool gfs2_log_try_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
0561               unsigned int *extra_revokes)
0562 {
0563     unsigned int blks = tr->tr_reserved;
0564     unsigned int revokes = tr->tr_revokes;
0565     unsigned int revoke_blks = 0;
0566 
0567     *extra_revokes = 0;
0568     if (revokes && !__gfs2_log_try_reserve_revokes(sdp, revokes)) {
0569         revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
0570         *extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
0571         blks += revoke_blks;
0572     }
0573     if (!blks)
0574         return true;
0575     if (__gfs2_log_try_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS))
0576         return true;
0577     if (!revoke_blks)
0578         gfs2_log_release_revokes(sdp, revokes);
0579     return false;
0580 }
0581 
0582 /**
0583  * gfs2_log_reserve - Make a log reservation
0584  * @sdp: The GFS2 superblock
0585  * @tr: The transaction
0586  * @extra_revokes: The number of additional revokes reserved (output)
0587  *
0588  * sdp->sd_log_flush_lock must not be held.
0589  */
0590 
0591 void gfs2_log_reserve(struct gfs2_sbd *sdp, struct gfs2_trans *tr,
0592               unsigned int *extra_revokes)
0593 {
0594     unsigned int blks = tr->tr_reserved;
0595     unsigned int revokes = tr->tr_revokes;
0596     unsigned int revoke_blks;
0597 
0598     *extra_revokes = 0;
0599     if (revokes) {
0600         revoke_blks = DIV_ROUND_UP(revokes, sdp->sd_inptrs);
0601         *extra_revokes = revoke_blks * sdp->sd_inptrs - revokes;
0602         blks += revoke_blks;
0603     }
0604     __gfs2_log_reserve(sdp, blks, GFS2_LOG_FLUSH_MIN_BLOCKS);
0605 }
0606 
0607 /**
0608  * log_distance - Compute distance between two journal blocks
0609  * @sdp: The GFS2 superblock
0610  * @newer: The most recent journal block of the pair
0611  * @older: The older journal block of the pair
0612  *
0613  *   Compute the distance (in the journal direction) between two
0614  *   blocks in the journal
0615  *
0616  * Returns: the distance in blocks
0617  */
0618 
0619 static inline unsigned int log_distance(struct gfs2_sbd *sdp, unsigned int newer,
0620                     unsigned int older)
0621 {
0622     int dist;
0623 
0624     dist = newer - older;
0625     if (dist < 0)
0626         dist += sdp->sd_jdesc->jd_blocks;
0627 
0628     return dist;
0629 }
0630 
0631 /**
0632  * calc_reserved - Calculate the number of blocks to keep reserved
0633  * @sdp: The GFS2 superblock
0634  *
0635  * This is complex.  We need to reserve room for all our currently used
0636  * metadata blocks (e.g. normal file I/O rewriting file time stamps) and
0637  * all our journaled data blocks for journaled files (e.g. files in the
0638  * meta_fs like rindex, or files for which chattr +j was done.)
0639  * If we don't reserve enough space, corruption will follow.
0640  *
0641  * We can have metadata blocks and jdata blocks in the same journal.  Each
0642  * type gets its own log descriptor, for which we need to reserve a block.
0643  * In fact, each type has the potential for needing more than one log descriptor
0644  * in cases where we have more blocks than will fit in a log descriptor.
0645  * Metadata journal entries take up half the space of journaled buffer entries.
0646  *
0647  * Also, we need to reserve blocks for revoke journal entries and one for an
0648  * overall header for the lot.
0649  *
0650  * Returns: the number of blocks reserved
0651  */
0652 static unsigned int calc_reserved(struct gfs2_sbd *sdp)
0653 {
0654     unsigned int reserved = GFS2_LOG_FLUSH_MIN_BLOCKS;
0655     unsigned int blocks;
0656     struct gfs2_trans *tr = sdp->sd_log_tr;
0657 
0658     if (tr) {
0659         blocks = tr->tr_num_buf_new - tr->tr_num_buf_rm;
0660         reserved += blocks + DIV_ROUND_UP(blocks, buf_limit(sdp));
0661         blocks = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
0662         reserved += blocks + DIV_ROUND_UP(blocks, databuf_limit(sdp));
0663     }
0664     return reserved;
0665 }
0666 
0667 static void log_pull_tail(struct gfs2_sbd *sdp)
0668 {
0669     unsigned int new_tail = sdp->sd_log_flush_tail;
0670     unsigned int dist;
0671 
0672     if (new_tail == sdp->sd_log_tail)
0673         return;
0674     dist = log_distance(sdp, new_tail, sdp->sd_log_tail);
0675     ail2_empty(sdp, new_tail);
0676     gfs2_log_release(sdp, dist);
0677     sdp->sd_log_tail = new_tail;
0678 }
0679 
0680 
0681 void log_flush_wait(struct gfs2_sbd *sdp)
0682 {
0683     DEFINE_WAIT(wait);
0684 
0685     if (atomic_read(&sdp->sd_log_in_flight)) {
0686         do {
0687             prepare_to_wait(&sdp->sd_log_flush_wait, &wait,
0688                     TASK_UNINTERRUPTIBLE);
0689             if (atomic_read(&sdp->sd_log_in_flight))
0690                 io_schedule();
0691         } while(atomic_read(&sdp->sd_log_in_flight));
0692         finish_wait(&sdp->sd_log_flush_wait, &wait);
0693     }
0694 }
0695 
0696 static int ip_cmp(void *priv, const struct list_head *a, const struct list_head *b)
0697 {
0698     struct gfs2_inode *ipa, *ipb;
0699 
0700     ipa = list_entry(a, struct gfs2_inode, i_ordered);
0701     ipb = list_entry(b, struct gfs2_inode, i_ordered);
0702 
0703     if (ipa->i_no_addr < ipb->i_no_addr)
0704         return -1;
0705     if (ipa->i_no_addr > ipb->i_no_addr)
0706         return 1;
0707     return 0;
0708 }
0709 
0710 static void __ordered_del_inode(struct gfs2_inode *ip)
0711 {
0712     if (!list_empty(&ip->i_ordered))
0713         list_del_init(&ip->i_ordered);
0714 }
0715 
0716 static void gfs2_ordered_write(struct gfs2_sbd *sdp)
0717 {
0718     struct gfs2_inode *ip;
0719     LIST_HEAD(written);
0720 
0721     spin_lock(&sdp->sd_ordered_lock);
0722     list_sort(NULL, &sdp->sd_log_ordered, &ip_cmp);
0723     while (!list_empty(&sdp->sd_log_ordered)) {
0724         ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
0725         if (ip->i_inode.i_mapping->nrpages == 0) {
0726             __ordered_del_inode(ip);
0727             continue;
0728         }
0729         list_move(&ip->i_ordered, &written);
0730         spin_unlock(&sdp->sd_ordered_lock);
0731         filemap_fdatawrite(ip->i_inode.i_mapping);
0732         spin_lock(&sdp->sd_ordered_lock);
0733     }
0734     list_splice(&written, &sdp->sd_log_ordered);
0735     spin_unlock(&sdp->sd_ordered_lock);
0736 }
0737 
0738 static void gfs2_ordered_wait(struct gfs2_sbd *sdp)
0739 {
0740     struct gfs2_inode *ip;
0741 
0742     spin_lock(&sdp->sd_ordered_lock);
0743     while (!list_empty(&sdp->sd_log_ordered)) {
0744         ip = list_first_entry(&sdp->sd_log_ordered, struct gfs2_inode, i_ordered);
0745         __ordered_del_inode(ip);
0746         if (ip->i_inode.i_mapping->nrpages == 0)
0747             continue;
0748         spin_unlock(&sdp->sd_ordered_lock);
0749         filemap_fdatawait(ip->i_inode.i_mapping);
0750         spin_lock(&sdp->sd_ordered_lock);
0751     }
0752     spin_unlock(&sdp->sd_ordered_lock);
0753 }
0754 
0755 void gfs2_ordered_del_inode(struct gfs2_inode *ip)
0756 {
0757     struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
0758 
0759     spin_lock(&sdp->sd_ordered_lock);
0760     __ordered_del_inode(ip);
0761     spin_unlock(&sdp->sd_ordered_lock);
0762 }
0763 
0764 void gfs2_add_revoke(struct gfs2_sbd *sdp, struct gfs2_bufdata *bd)
0765 {
0766     struct buffer_head *bh = bd->bd_bh;
0767     struct gfs2_glock *gl = bd->bd_gl;
0768 
0769     sdp->sd_log_num_revoke++;
0770     if (atomic_inc_return(&gl->gl_revokes) == 1)
0771         gfs2_glock_hold(gl);
0772     bh->b_private = NULL;
0773     bd->bd_blkno = bh->b_blocknr;
0774     gfs2_remove_from_ail(bd); /* drops ref on bh */
0775     bd->bd_bh = NULL;
0776     set_bit(GLF_LFLUSH, &gl->gl_flags);
0777     list_add(&bd->bd_list, &sdp->sd_log_revokes);
0778 }
0779 
0780 void gfs2_glock_remove_revoke(struct gfs2_glock *gl)
0781 {
0782     if (atomic_dec_return(&gl->gl_revokes) == 0) {
0783         clear_bit(GLF_LFLUSH, &gl->gl_flags);
0784         gfs2_glock_queue_put(gl);
0785     }
0786 }
0787 
0788 /**
0789  * gfs2_flush_revokes - Add as many revokes to the system transaction as we can
0790  * @sdp: The GFS2 superblock
0791  *
0792  * Our usual strategy is to defer writing revokes as much as we can in the hope
0793  * that we'll eventually overwrite the journal, which will make those revokes
0794  * go away.  This changes when we flush the log: at that point, there will
0795  * likely be some left-over space in the last revoke block of that transaction.
0796  * We can fill that space with additional revokes for blocks that have already
0797  * been written back.  This will basically come at no cost now, and will save
0798  * us from having to keep track of those blocks on the AIL2 list later.
0799  */
0800 void gfs2_flush_revokes(struct gfs2_sbd *sdp)
0801 {
0802     /* number of revokes we still have room for */
0803     unsigned int max_revokes = atomic_read(&sdp->sd_log_revokes_available);
0804 
0805     gfs2_log_lock(sdp);
0806     gfs2_ail1_empty(sdp, max_revokes);
0807     gfs2_log_unlock(sdp);
0808 }
0809 
0810 /**
0811  * gfs2_write_log_header - Write a journal log header buffer at lblock
0812  * @sdp: The GFS2 superblock
0813  * @jd: journal descriptor of the journal to which we are writing
0814  * @seq: sequence number
0815  * @tail: tail of the log
0816  * @lblock: value for lh_blkno (block number relative to start of journal)
0817  * @flags: log header flags GFS2_LOG_HEAD_*
0818  * @op_flags: flags to pass to the bio
0819  *
0820  * Returns: the initialized log buffer descriptor
0821  */
0822 
0823 void gfs2_write_log_header(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
0824                u64 seq, u32 tail, u32 lblock, u32 flags,
0825                blk_opf_t op_flags)
0826 {
0827     struct gfs2_log_header *lh;
0828     u32 hash, crc;
0829     struct page *page;
0830     struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
0831     struct timespec64 tv;
0832     struct super_block *sb = sdp->sd_vfs;
0833     u64 dblock;
0834 
0835     if (gfs2_withdrawn(sdp))
0836         return;
0837 
0838     page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
0839     lh = page_address(page);
0840     clear_page(lh);
0841 
0842     lh->lh_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
0843     lh->lh_header.mh_type = cpu_to_be32(GFS2_METATYPE_LH);
0844     lh->lh_header.__pad0 = cpu_to_be64(0);
0845     lh->lh_header.mh_format = cpu_to_be32(GFS2_FORMAT_LH);
0846     lh->lh_header.mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
0847     lh->lh_sequence = cpu_to_be64(seq);
0848     lh->lh_flags = cpu_to_be32(flags);
0849     lh->lh_tail = cpu_to_be32(tail);
0850     lh->lh_blkno = cpu_to_be32(lblock);
0851     hash = ~crc32(~0, lh, LH_V1_SIZE);
0852     lh->lh_hash = cpu_to_be32(hash);
0853 
0854     ktime_get_coarse_real_ts64(&tv);
0855     lh->lh_nsec = cpu_to_be32(tv.tv_nsec);
0856     lh->lh_sec = cpu_to_be64(tv.tv_sec);
0857     if (!list_empty(&jd->extent_list))
0858         dblock = gfs2_log_bmap(jd, lblock);
0859     else {
0860         unsigned int extlen;
0861         int ret;
0862 
0863         extlen = 1;
0864         ret = gfs2_get_extent(jd->jd_inode, lblock, &dblock, &extlen);
0865         if (gfs2_assert_withdraw(sdp, ret == 0))
0866             return;
0867     }
0868     lh->lh_addr = cpu_to_be64(dblock);
0869     lh->lh_jinode = cpu_to_be64(GFS2_I(jd->jd_inode)->i_no_addr);
0870 
0871     /* We may only write local statfs, quota, etc., when writing to our
0872        own journal. The values are left 0 when recovering a journal
0873        different from our own. */
0874     if (!(flags & GFS2_LOG_HEAD_RECOVERY)) {
0875         lh->lh_statfs_addr =
0876             cpu_to_be64(GFS2_I(sdp->sd_sc_inode)->i_no_addr);
0877         lh->lh_quota_addr =
0878             cpu_to_be64(GFS2_I(sdp->sd_qc_inode)->i_no_addr);
0879 
0880         spin_lock(&sdp->sd_statfs_spin);
0881         lh->lh_local_total = cpu_to_be64(l_sc->sc_total);
0882         lh->lh_local_free = cpu_to_be64(l_sc->sc_free);
0883         lh->lh_local_dinodes = cpu_to_be64(l_sc->sc_dinodes);
0884         spin_unlock(&sdp->sd_statfs_spin);
0885     }
0886 
0887     BUILD_BUG_ON(offsetof(struct gfs2_log_header, lh_crc) != LH_V1_SIZE);
0888 
0889     crc = crc32c(~0, (void *)lh + LH_V1_SIZE + 4,
0890              sb->s_blocksize - LH_V1_SIZE - 4);
0891     lh->lh_crc = cpu_to_be32(crc);
0892 
0893     gfs2_log_write(sdp, jd, page, sb->s_blocksize, 0, dblock);
0894     gfs2_log_submit_bio(&jd->jd_log_bio, REQ_OP_WRITE | op_flags);
0895 }
0896 
0897 /**
0898  * log_write_header - Get and initialize a journal header buffer
0899  * @sdp: The GFS2 superblock
0900  * @flags: The log header flags, including log header origin
0901  *
0902  * Returns: the initialized log buffer descriptor
0903  */
0904 
0905 static void log_write_header(struct gfs2_sbd *sdp, u32 flags)
0906 {
0907     blk_opf_t op_flags = REQ_PREFLUSH | REQ_FUA | REQ_META | REQ_SYNC;
0908     enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
0909 
0910     gfs2_assert_withdraw(sdp, (state != SFS_FROZEN));
0911 
0912     if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags)) {
0913         gfs2_ordered_wait(sdp);
0914         log_flush_wait(sdp);
0915         op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
0916     }
0917     sdp->sd_log_idle = (sdp->sd_log_flush_tail == sdp->sd_log_flush_head);
0918     gfs2_write_log_header(sdp, sdp->sd_jdesc, sdp->sd_log_sequence++,
0919                   sdp->sd_log_flush_tail, sdp->sd_log_flush_head,
0920                   flags, op_flags);
0921     gfs2_log_incr_head(sdp);
0922     log_flush_wait(sdp);
0923     log_pull_tail(sdp);
0924     gfs2_log_update_head(sdp);
0925 }
0926 
0927 /**
0928  * gfs2_ail_drain - drain the ail lists after a withdraw
0929  * @sdp: Pointer to GFS2 superblock
0930  */
0931 void gfs2_ail_drain(struct gfs2_sbd *sdp)
0932 {
0933     struct gfs2_trans *tr;
0934 
0935     spin_lock(&sdp->sd_ail_lock);
0936     /*
0937      * For transactions on the sd_ail1_list we need to drain both the
0938      * ail1 and ail2 lists. That's because function gfs2_ail1_start_one
0939      * (temporarily) moves items from its tr_ail1 list to tr_ail2 list
0940      * before revokes are sent for that block. Items on the sd_ail2_list
0941      * should have already gotten beyond that point, so no need.
0942      */
0943     while (!list_empty(&sdp->sd_ail1_list)) {
0944         tr = list_first_entry(&sdp->sd_ail1_list, struct gfs2_trans,
0945                       tr_list);
0946         gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail1_list);
0947         gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
0948         list_del(&tr->tr_list);
0949         gfs2_trans_free(sdp, tr);
0950     }
0951     while (!list_empty(&sdp->sd_ail2_list)) {
0952         tr = list_first_entry(&sdp->sd_ail2_list, struct gfs2_trans,
0953                       tr_list);
0954         gfs2_ail_empty_tr(sdp, tr, &tr->tr_ail2_list);
0955         list_del(&tr->tr_list);
0956         gfs2_trans_free(sdp, tr);
0957     }
0958     gfs2_drain_revokes(sdp);
0959     spin_unlock(&sdp->sd_ail_lock);
0960 }
0961 
0962 /**
0963  * empty_ail1_list - try to start IO and empty the ail1 list
0964  * @sdp: Pointer to GFS2 superblock
0965  */
0966 static void empty_ail1_list(struct gfs2_sbd *sdp)
0967 {
0968     unsigned long start = jiffies;
0969 
0970     for (;;) {
0971         if (time_after(jiffies, start + (HZ * 600))) {
0972             fs_err(sdp, "Error: In %s for 10 minutes! t=%d\n",
0973                    __func__, current->journal_info ? 1 : 0);
0974             dump_ail_list(sdp);
0975             return;
0976         }
0977         gfs2_ail1_start(sdp);
0978         gfs2_ail1_wait(sdp);
0979         if (gfs2_ail1_empty(sdp, 0))
0980             return;
0981     }
0982 }
0983 
0984 /**
0985  * trans_drain - drain the buf and databuf queue for a failed transaction
0986  * @tr: the transaction to drain
0987  *
0988  * When this is called, we're taking an error exit for a log write that failed
0989  * but since we bypassed the after_commit functions, we need to remove the
0990  * items from the buf and databuf queue.
0991  */
0992 static void trans_drain(struct gfs2_trans *tr)
0993 {
0994     struct gfs2_bufdata *bd;
0995     struct list_head *head;
0996 
0997     if (!tr)
0998         return;
0999 
1000     head = &tr->tr_buf;
1001     while (!list_empty(head)) {
1002         bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1003         list_del_init(&bd->bd_list);
1004         if (!list_empty(&bd->bd_ail_st_list))
1005             gfs2_remove_from_ail(bd);
1006         kmem_cache_free(gfs2_bufdata_cachep, bd);
1007     }
1008     head = &tr->tr_databuf;
1009     while (!list_empty(head)) {
1010         bd = list_first_entry(head, struct gfs2_bufdata, bd_list);
1011         list_del_init(&bd->bd_list);
1012         if (!list_empty(&bd->bd_ail_st_list))
1013             gfs2_remove_from_ail(bd);
1014         kmem_cache_free(gfs2_bufdata_cachep, bd);
1015     }
1016 }
1017 
1018 /**
1019  * gfs2_log_flush - flush incore transaction(s)
1020  * @sdp: The filesystem
1021  * @gl: The glock structure to flush.  If NULL, flush the whole incore log
1022  * @flags: The log header flags: GFS2_LOG_HEAD_FLUSH_* and debug flags
1023  *
1024  */
1025 
1026 void gfs2_log_flush(struct gfs2_sbd *sdp, struct gfs2_glock *gl, u32 flags)
1027 {
1028     struct gfs2_trans *tr = NULL;
1029     unsigned int reserved_blocks = 0, used_blocks = 0;
1030     enum gfs2_freeze_state state = atomic_read(&sdp->sd_freeze_state);
1031     unsigned int first_log_head;
1032     unsigned int reserved_revokes = 0;
1033 
1034     down_write(&sdp->sd_log_flush_lock);
1035     trace_gfs2_log_flush(sdp, 1, flags);
1036 
1037 repeat:
1038     /*
1039      * Do this check while holding the log_flush_lock to prevent new
1040      * buffers from being added to the ail via gfs2_pin()
1041      */
1042     if (gfs2_withdrawn(sdp) || !test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
1043         goto out;
1044 
1045     /* Log might have been flushed while we waited for the flush lock */
1046     if (gl && !test_bit(GLF_LFLUSH, &gl->gl_flags))
1047         goto out;
1048 
1049     first_log_head = sdp->sd_log_head;
1050     sdp->sd_log_flush_head = first_log_head;
1051 
1052     tr = sdp->sd_log_tr;
1053     if (tr || sdp->sd_log_num_revoke) {
1054         if (reserved_blocks)
1055             gfs2_log_release(sdp, reserved_blocks);
1056         reserved_blocks = sdp->sd_log_blks_reserved;
1057         reserved_revokes = sdp->sd_log_num_revoke;
1058         if (tr) {
1059             sdp->sd_log_tr = NULL;
1060             tr->tr_first = first_log_head;
1061             if (unlikely (state == SFS_FROZEN)) {
1062                 if (gfs2_assert_withdraw_delayed(sdp,
1063                        !tr->tr_num_buf_new && !tr->tr_num_databuf_new))
1064                     goto out_withdraw;
1065             }
1066         }
1067     } else if (!reserved_blocks) {
1068         unsigned int taboo_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
1069 
1070         reserved_blocks = GFS2_LOG_FLUSH_MIN_BLOCKS;
1071         if (current == sdp->sd_logd_process)
1072             taboo_blocks = 0;
1073 
1074         if (!__gfs2_log_try_reserve(sdp, reserved_blocks, taboo_blocks)) {
1075             up_write(&sdp->sd_log_flush_lock);
1076             __gfs2_log_reserve(sdp, reserved_blocks, taboo_blocks);
1077             down_write(&sdp->sd_log_flush_lock);
1078             goto repeat;
1079         }
1080         BUG_ON(sdp->sd_log_num_revoke);
1081     }
1082 
1083     if (flags & GFS2_LOG_HEAD_FLUSH_SHUTDOWN)
1084         clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
1085 
1086     if (unlikely(state == SFS_FROZEN))
1087         if (gfs2_assert_withdraw_delayed(sdp, !reserved_revokes))
1088             goto out_withdraw;
1089 
1090     gfs2_ordered_write(sdp);
1091     if (gfs2_withdrawn(sdp))
1092         goto out_withdraw;
1093     lops_before_commit(sdp, tr);
1094     if (gfs2_withdrawn(sdp))
1095         goto out_withdraw;
1096     gfs2_log_submit_bio(&sdp->sd_jdesc->jd_log_bio, REQ_OP_WRITE);
1097     if (gfs2_withdrawn(sdp))
1098         goto out_withdraw;
1099 
1100     if (sdp->sd_log_head != sdp->sd_log_flush_head) {
1101         log_write_header(sdp, flags);
1102     } else if (sdp->sd_log_tail != sdp->sd_log_flush_tail && !sdp->sd_log_idle) {
1103         log_write_header(sdp, flags);
1104     }
1105     if (gfs2_withdrawn(sdp))
1106         goto out_withdraw;
1107     lops_after_commit(sdp, tr);
1108 
1109     gfs2_log_lock(sdp);
1110     sdp->sd_log_blks_reserved = 0;
1111 
1112     spin_lock(&sdp->sd_ail_lock);
1113     if (tr && !list_empty(&tr->tr_ail1_list)) {
1114         list_add(&tr->tr_list, &sdp->sd_ail1_list);
1115         tr = NULL;
1116     }
1117     spin_unlock(&sdp->sd_ail_lock);
1118     gfs2_log_unlock(sdp);
1119 
1120     if (!(flags & GFS2_LOG_HEAD_FLUSH_NORMAL)) {
1121         if (!sdp->sd_log_idle) {
1122             empty_ail1_list(sdp);
1123             if (gfs2_withdrawn(sdp))
1124                 goto out_withdraw;
1125             log_write_header(sdp, flags);
1126         }
1127         if (flags & (GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
1128                  GFS2_LOG_HEAD_FLUSH_FREEZE))
1129             gfs2_log_shutdown(sdp);
1130         if (flags & GFS2_LOG_HEAD_FLUSH_FREEZE)
1131             atomic_set(&sdp->sd_freeze_state, SFS_FROZEN);
1132     }
1133 
1134 out_end:
1135     used_blocks = log_distance(sdp, sdp->sd_log_flush_head, first_log_head);
1136     reserved_revokes += atomic_read(&sdp->sd_log_revokes_available);
1137     atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
1138     gfs2_assert_withdraw(sdp, reserved_revokes % sdp->sd_inptrs == sdp->sd_ldptrs);
1139     if (reserved_revokes > sdp->sd_ldptrs)
1140         reserved_blocks += (reserved_revokes - sdp->sd_ldptrs) / sdp->sd_inptrs;
1141 out:
1142     if (used_blocks != reserved_blocks) {
1143         gfs2_assert_withdraw_delayed(sdp, used_blocks < reserved_blocks);
1144         gfs2_log_release(sdp, reserved_blocks - used_blocks);
1145     }
1146     up_write(&sdp->sd_log_flush_lock);
1147     gfs2_trans_free(sdp, tr);
1148     if (gfs2_withdrawing(sdp))
1149         gfs2_withdraw(sdp);
1150     trace_gfs2_log_flush(sdp, 0, flags);
1151     return;
1152 
1153 out_withdraw:
1154     trans_drain(tr);
1155     /**
1156      * If the tr_list is empty, we're withdrawing during a log
1157      * flush that targets a transaction, but the transaction was
1158      * never queued onto any of the ail lists. Here we add it to
1159      * ail1 just so that ail_drain() will find and free it.
1160      */
1161     spin_lock(&sdp->sd_ail_lock);
1162     if (tr && list_empty(&tr->tr_list))
1163         list_add(&tr->tr_list, &sdp->sd_ail1_list);
1164     spin_unlock(&sdp->sd_ail_lock);
1165     tr = NULL;
1166     goto out_end;
1167 }
1168 
1169 /**
1170  * gfs2_merge_trans - Merge a new transaction into a cached transaction
1171  * @sdp: the filesystem
1172  * @new: New transaction to be merged
1173  */
1174 
1175 static void gfs2_merge_trans(struct gfs2_sbd *sdp, struct gfs2_trans *new)
1176 {
1177     struct gfs2_trans *old = sdp->sd_log_tr;
1178 
1179     WARN_ON_ONCE(!test_bit(TR_ATTACHED, &old->tr_flags));
1180 
1181     old->tr_num_buf_new += new->tr_num_buf_new;
1182     old->tr_num_databuf_new += new->tr_num_databuf_new;
1183     old->tr_num_buf_rm  += new->tr_num_buf_rm;
1184     old->tr_num_databuf_rm  += new->tr_num_databuf_rm;
1185     old->tr_revokes     += new->tr_revokes;
1186     old->tr_num_revoke  += new->tr_num_revoke;
1187 
1188     list_splice_tail_init(&new->tr_databuf, &old->tr_databuf);
1189     list_splice_tail_init(&new->tr_buf, &old->tr_buf);
1190 
1191     spin_lock(&sdp->sd_ail_lock);
1192     list_splice_tail_init(&new->tr_ail1_list, &old->tr_ail1_list);
1193     list_splice_tail_init(&new->tr_ail2_list, &old->tr_ail2_list);
1194     spin_unlock(&sdp->sd_ail_lock);
1195 }
1196 
1197 static void log_refund(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1198 {
1199     unsigned int reserved;
1200     unsigned int unused;
1201     unsigned int maxres;
1202 
1203     gfs2_log_lock(sdp);
1204 
1205     if (sdp->sd_log_tr) {
1206         gfs2_merge_trans(sdp, tr);
1207     } else if (tr->tr_num_buf_new || tr->tr_num_databuf_new) {
1208         gfs2_assert_withdraw(sdp, !test_bit(TR_ONSTACK, &tr->tr_flags));
1209         sdp->sd_log_tr = tr;
1210         set_bit(TR_ATTACHED, &tr->tr_flags);
1211     }
1212 
1213     reserved = calc_reserved(sdp);
1214     maxres = sdp->sd_log_blks_reserved + tr->tr_reserved;
1215     gfs2_assert_withdraw(sdp, maxres >= reserved);
1216     unused = maxres - reserved;
1217     if (unused)
1218         gfs2_log_release(sdp, unused);
1219     sdp->sd_log_blks_reserved = reserved;
1220 
1221     gfs2_log_unlock(sdp);
1222 }
1223 
1224 /**
1225  * gfs2_log_commit - Commit a transaction to the log
1226  * @sdp: the filesystem
1227  * @tr: the transaction
1228  *
1229  * We wake up gfs2_logd if the number of pinned blocks exceed thresh1
1230  * or the total number of used blocks (pinned blocks plus AIL blocks)
1231  * is greater than thresh2.
1232  *
1233  * At mount time thresh1 is 2/5ths of journal size, thresh2 is 4/5ths of
1234  * journal size.
1235  *
1236  * Returns: errno
1237  */
1238 
1239 void gfs2_log_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1240 {
1241     log_refund(sdp, tr);
1242 
1243     if (atomic_read(&sdp->sd_log_pinned) > atomic_read(&sdp->sd_log_thresh1) ||
1244         ((sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free)) >
1245         atomic_read(&sdp->sd_log_thresh2)))
1246         wake_up(&sdp->sd_logd_waitq);
1247 }
1248 
1249 /**
1250  * gfs2_log_shutdown - write a shutdown header into a journal
1251  * @sdp: the filesystem
1252  *
1253  */
1254 
1255 static void gfs2_log_shutdown(struct gfs2_sbd *sdp)
1256 {
1257     gfs2_assert_withdraw(sdp, !sdp->sd_log_blks_reserved);
1258     gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
1259     gfs2_assert_withdraw(sdp, list_empty(&sdp->sd_ail1_list));
1260 
1261     log_write_header(sdp, GFS2_LOG_HEAD_UNMOUNT | GFS2_LFC_SHUTDOWN);
1262     log_pull_tail(sdp);
1263 
1264     gfs2_assert_warn(sdp, sdp->sd_log_head == sdp->sd_log_tail);
1265     gfs2_assert_warn(sdp, list_empty(&sdp->sd_ail2_list));
1266 }
1267 
1268 static inline int gfs2_jrnl_flush_reqd(struct gfs2_sbd *sdp)
1269 {
1270     return (atomic_read(&sdp->sd_log_pinned) +
1271         atomic_read(&sdp->sd_log_blks_needed) >=
1272         atomic_read(&sdp->sd_log_thresh1));
1273 }
1274 
1275 static inline int gfs2_ail_flush_reqd(struct gfs2_sbd *sdp)
1276 {
1277     unsigned int used_blocks = sdp->sd_jdesc->jd_blocks - atomic_read(&sdp->sd_log_blks_free);
1278 
1279     if (test_and_clear_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags))
1280         return 1;
1281 
1282     return used_blocks + atomic_read(&sdp->sd_log_blks_needed) >=
1283         atomic_read(&sdp->sd_log_thresh2);
1284 }
1285 
1286 /**
1287  * gfs2_logd - Update log tail as Active Items get flushed to in-place blocks
1288  * @data: Pointer to GFS2 superblock
1289  *
1290  * Also, periodically check to make sure that we're using the most recent
1291  * journal index.
1292  */
1293 
1294 int gfs2_logd(void *data)
1295 {
1296     struct gfs2_sbd *sdp = data;
1297     unsigned long t = 1;
1298     DEFINE_WAIT(wait);
1299 
1300     while (!kthread_should_stop()) {
1301 
1302         if (gfs2_withdrawn(sdp)) {
1303             msleep_interruptible(HZ);
1304             continue;
1305         }
1306         /* Check for errors writing to the journal */
1307         if (sdp->sd_log_error) {
1308             gfs2_lm(sdp,
1309                 "GFS2: fsid=%s: error %d: "
1310                 "withdrawing the file system to "
1311                 "prevent further damage.\n",
1312                 sdp->sd_fsname, sdp->sd_log_error);
1313             gfs2_withdraw(sdp);
1314             continue;
1315         }
1316 
1317         if (gfs2_jrnl_flush_reqd(sdp) || t == 0) {
1318             gfs2_ail1_empty(sdp, 0);
1319             gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1320                           GFS2_LFC_LOGD_JFLUSH_REQD);
1321         }
1322 
1323         if (gfs2_ail_flush_reqd(sdp)) {
1324             gfs2_ail1_start(sdp);
1325             gfs2_ail1_wait(sdp);
1326             gfs2_ail1_empty(sdp, 0);
1327             gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
1328                           GFS2_LFC_LOGD_AIL_FLUSH_REQD);
1329         }
1330 
1331         t = gfs2_tune_get(sdp, gt_logd_secs) * HZ;
1332 
1333         try_to_freeze();
1334 
1335         do {
1336             prepare_to_wait(&sdp->sd_logd_waitq, &wait,
1337                     TASK_INTERRUPTIBLE);
1338             if (!gfs2_ail_flush_reqd(sdp) &&
1339                 !gfs2_jrnl_flush_reqd(sdp) &&
1340                 !kthread_should_stop())
1341                 t = schedule_timeout(t);
1342         } while(t && !gfs2_ail_flush_reqd(sdp) &&
1343             !gfs2_jrnl_flush_reqd(sdp) &&
1344             !kthread_should_stop());
1345         finish_wait(&sdp->sd_logd_waitq, &wait);
1346     }
1347 
1348     return 0;
1349 }
1350
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!