Back to home page

OSCL-LXR
 
 

     

0001 /*
0002  * Copyright (C) 2014 Facebook. All rights reserved.
0003  *
0004  * This file is released under the GPL.
0005  */
0006 
0007 #include <linux/device-mapper.h>
0008 
0009 #include <linux/module.h>
0010 #include <linux/init.h>
0011 #include <linux/blkdev.h>
0012 #include <linux/bio.h>
0013 #include <linux/dax.h>
0014 #include <linux/slab.h>
0015 #include <linux/kthread.h>
0016 #include <linux/freezer.h>
0017 #include <linux/uio.h>
0018 
0019 #define DM_MSG_PREFIX "log-writes"
0020 
0021 /*
0022  * This target will sequentially log all writes to the target device onto the
0023  * log device.  This is helpful for replaying writes to check for fs consistency
0024  * at all times.  This target provides a mechanism to mark specific events to
0025  * check data at a later time.  So for example you would:
0026  *
0027  * write data
0028  * fsync
0029  * dmsetup message /dev/whatever mark mymark
0030  * unmount /mnt/test
0031  *
0032  * Then replay the log up to mymark and check the contents of the replay to
0033  * verify it matches what was written.
0034  *
0035  * We log writes only after they have been flushed, this makes the log describe
0036  * close to the order in which the data hits the actual disk, not its cache.  So
0037  * for example the following sequence (W means write, C means complete)
0038  *
0039  * Wa,Wb,Wc,Cc,Ca,FLUSH,FUAd,Cb,CFLUSH,CFUAd
0040  *
0041  * Would result in the log looking like this:
0042  *
0043  * c,a,b,flush,fuad,<other writes>,<next flush>
0044  *
0045  * This is meant to help expose problems where file systems do not properly wait
0046  * on data being written before invoking a FLUSH.  FUA bypasses cache so once it
0047  * completes it is added to the log as it should be on disk.
0048  *
0049  * We treat DISCARDs as if they don't bypass cache so that they are logged in
0050  * order of completion along with the normal writes.  If we didn't do it this
0051  * way we would process all the discards first and then write all the data, when
0052  * in fact we want to do the data and the discard in the order that they
0053  * completed.
0054  */
0055 #define LOG_FLUSH_FLAG      (1 << 0)
0056 #define LOG_FUA_FLAG        (1 << 1)
0057 #define LOG_DISCARD_FLAG    (1 << 2)
0058 #define LOG_MARK_FLAG       (1 << 3)
0059 #define LOG_METADATA_FLAG   (1 << 4)
0060 
0061 #define WRITE_LOG_VERSION 1ULL
0062 #define WRITE_LOG_MAGIC 0x6a736677736872ULL
0063 #define WRITE_LOG_SUPER_SECTOR 0
0064 
0065 /*
0066  * The disk format for this is braindead simple.
0067  *
0068  * At byte 0 we have our super, followed by the following sequence for
0069  * nr_entries:
0070  *
0071  * [   1 sector    ][  entry->nr_sectors ]
0072  * [log_write_entry][    data written    ]
0073  *
0074  * The log_write_entry takes up a full sector so we can have arbitrary length
0075  * marks and it leaves us room for extra content in the future.
0076  */
0077 
0078 /*
0079  * Basic info about the log for userspace.
0080  */
0081 struct log_write_super {
0082     __le64 magic;
0083     __le64 version;
0084     __le64 nr_entries;
0085     __le32 sectorsize;
0086 };
0087 
0088 /*
0089  * sector - the sector we wrote.
0090  * nr_sectors - the number of sectors we wrote.
0091  * flags - flags for this log entry.
0092  * data_len - the size of the data in this log entry, this is for private log
0093  * entry stuff, the MARK data provided by userspace for example.
0094  */
0095 struct log_write_entry {
0096     __le64 sector;
0097     __le64 nr_sectors;
0098     __le64 flags;
0099     __le64 data_len;
0100 };
0101 
0102 struct log_writes_c {
0103     struct dm_dev *dev;
0104     struct dm_dev *logdev;
0105     u64 logged_entries;
0106     u32 sectorsize;
0107     u32 sectorshift;
0108     atomic_t io_blocks;
0109     atomic_t pending_blocks;
0110     sector_t next_sector;
0111     sector_t end_sector;
0112     bool logging_enabled;
0113     bool device_supports_discard;
0114     spinlock_t blocks_lock;
0115     struct list_head unflushed_blocks;
0116     struct list_head logging_blocks;
0117     wait_queue_head_t wait;
0118     struct task_struct *log_kthread;
0119     struct completion super_done;
0120 };
0121 
0122 struct pending_block {
0123     int vec_cnt;
0124     u64 flags;
0125     sector_t sector;
0126     sector_t nr_sectors;
0127     char *data;
0128     u32 datalen;
0129     struct list_head list;
0130     struct bio_vec vecs[];
0131 };
0132 
0133 struct per_bio_data {
0134     struct pending_block *block;
0135 };
0136 
0137 static inline sector_t bio_to_dev_sectors(struct log_writes_c *lc,
0138                       sector_t sectors)
0139 {
0140     return sectors >> (lc->sectorshift - SECTOR_SHIFT);
0141 }
0142 
0143 static inline sector_t dev_to_bio_sectors(struct log_writes_c *lc,
0144                       sector_t sectors)
0145 {
0146     return sectors << (lc->sectorshift - SECTOR_SHIFT);
0147 }
0148 
0149 static void put_pending_block(struct log_writes_c *lc)
0150 {
0151     if (atomic_dec_and_test(&lc->pending_blocks)) {
0152         smp_mb__after_atomic();
0153         if (waitqueue_active(&lc->wait))
0154             wake_up(&lc->wait);
0155     }
0156 }
0157 
0158 static void put_io_block(struct log_writes_c *lc)
0159 {
0160     if (atomic_dec_and_test(&lc->io_blocks)) {
0161         smp_mb__after_atomic();
0162         if (waitqueue_active(&lc->wait))
0163             wake_up(&lc->wait);
0164     }
0165 }
0166 
0167 static void log_end_io(struct bio *bio)
0168 {
0169     struct log_writes_c *lc = bio->bi_private;
0170 
0171     if (bio->bi_status) {
0172         unsigned long flags;
0173 
0174         DMERR("Error writing log block, error=%d", bio->bi_status);
0175         spin_lock_irqsave(&lc->blocks_lock, flags);
0176         lc->logging_enabled = false;
0177         spin_unlock_irqrestore(&lc->blocks_lock, flags);
0178     }
0179 
0180     bio_free_pages(bio);
0181     put_io_block(lc);
0182     bio_put(bio);
0183 }
0184 
0185 static void log_end_super(struct bio *bio)
0186 {
0187     struct log_writes_c *lc = bio->bi_private;
0188 
0189     complete(&lc->super_done);
0190     log_end_io(bio);
0191 }
0192 
0193 /*
0194  * Meant to be called if there is an error, it will free all the pages
0195  * associated with the block.
0196  */
0197 static void free_pending_block(struct log_writes_c *lc,
0198                    struct pending_block *block)
0199 {
0200     int i;
0201 
0202     for (i = 0; i < block->vec_cnt; i++) {
0203         if (block->vecs[i].bv_page)
0204             __free_page(block->vecs[i].bv_page);
0205     }
0206     kfree(block->data);
0207     kfree(block);
0208     put_pending_block(lc);
0209 }
0210 
0211 static int write_metadata(struct log_writes_c *lc, void *entry,
0212               size_t entrylen, void *data, size_t datalen,
0213               sector_t sector)
0214 {
0215     struct bio *bio;
0216     struct page *page;
0217     void *ptr;
0218     size_t ret;
0219 
0220     bio = bio_alloc(lc->logdev->bdev, 1, REQ_OP_WRITE, GFP_KERNEL);
0221     bio->bi_iter.bi_size = 0;
0222     bio->bi_iter.bi_sector = sector;
0223     bio->bi_end_io = (sector == WRITE_LOG_SUPER_SECTOR) ?
0224               log_end_super : log_end_io;
0225     bio->bi_private = lc;
0226 
0227     page = alloc_page(GFP_KERNEL);
0228     if (!page) {
0229         DMERR("Couldn't alloc log page");
0230         bio_put(bio);
0231         goto error;
0232     }
0233 
0234     ptr = kmap_atomic(page);
0235     memcpy(ptr, entry, entrylen);
0236     if (datalen)
0237         memcpy(ptr + entrylen, data, datalen);
0238     memset(ptr + entrylen + datalen, 0,
0239            lc->sectorsize - entrylen - datalen);
0240     kunmap_atomic(ptr);
0241 
0242     ret = bio_add_page(bio, page, lc->sectorsize, 0);
0243     if (ret != lc->sectorsize) {
0244         DMERR("Couldn't add page to the log block");
0245         goto error_bio;
0246     }
0247     submit_bio(bio);
0248     return 0;
0249 error_bio:
0250     bio_put(bio);
0251     __free_page(page);
0252 error:
0253     put_io_block(lc);
0254     return -1;
0255 }
0256 
0257 static int write_inline_data(struct log_writes_c *lc, void *entry,
0258                  size_t entrylen, void *data, size_t datalen,
0259                  sector_t sector)
0260 {
0261     int bio_pages, pg_datalen, pg_sectorlen, i;
0262     struct page *page;
0263     struct bio *bio;
0264     size_t ret;
0265     void *ptr;
0266 
0267     while (datalen) {
0268         bio_pages = bio_max_segs(DIV_ROUND_UP(datalen, PAGE_SIZE));
0269 
0270         atomic_inc(&lc->io_blocks);
0271 
0272         bio = bio_alloc(lc->logdev->bdev, bio_pages, REQ_OP_WRITE,
0273                 GFP_KERNEL);
0274         bio->bi_iter.bi_size = 0;
0275         bio->bi_iter.bi_sector = sector;
0276         bio->bi_end_io = log_end_io;
0277         bio->bi_private = lc;
0278 
0279         for (i = 0; i < bio_pages; i++) {
0280             pg_datalen = min_t(int, datalen, PAGE_SIZE);
0281             pg_sectorlen = ALIGN(pg_datalen, lc->sectorsize);
0282 
0283             page = alloc_page(GFP_KERNEL);
0284             if (!page) {
0285                 DMERR("Couldn't alloc inline data page");
0286                 goto error_bio;
0287             }
0288 
0289             ptr = kmap_atomic(page);
0290             memcpy(ptr, data, pg_datalen);
0291             if (pg_sectorlen > pg_datalen)
0292                 memset(ptr + pg_datalen, 0, pg_sectorlen - pg_datalen);
0293             kunmap_atomic(ptr);
0294 
0295             ret = bio_add_page(bio, page, pg_sectorlen, 0);
0296             if (ret != pg_sectorlen) {
0297                 DMERR("Couldn't add page of inline data");
0298                 __free_page(page);
0299                 goto error_bio;
0300             }
0301 
0302             datalen -= pg_datalen;
0303             data    += pg_datalen;
0304         }
0305         submit_bio(bio);
0306 
0307         sector += bio_pages * PAGE_SECTORS;
0308     }
0309     return 0;
0310 error_bio:
0311     bio_free_pages(bio);
0312     bio_put(bio);
0313     put_io_block(lc);
0314     return -1;
0315 }
0316 
0317 static int log_one_block(struct log_writes_c *lc,
0318              struct pending_block *block, sector_t sector)
0319 {
0320     struct bio *bio;
0321     struct log_write_entry entry;
0322     size_t metadatalen, ret;
0323     int i;
0324 
0325     entry.sector = cpu_to_le64(block->sector);
0326     entry.nr_sectors = cpu_to_le64(block->nr_sectors);
0327     entry.flags = cpu_to_le64(block->flags);
0328     entry.data_len = cpu_to_le64(block->datalen);
0329 
0330     metadatalen = (block->flags & LOG_MARK_FLAG) ? block->datalen : 0;
0331     if (write_metadata(lc, &entry, sizeof(entry), block->data,
0332                metadatalen, sector)) {
0333         free_pending_block(lc, block);
0334         return -1;
0335     }
0336 
0337     sector += dev_to_bio_sectors(lc, 1);
0338 
0339     if (block->datalen && metadatalen == 0) {
0340         if (write_inline_data(lc, &entry, sizeof(entry), block->data,
0341                       block->datalen, sector)) {
0342             free_pending_block(lc, block);
0343             return -1;
0344         }
0345         /* we don't support both inline data & bio data */
0346         goto out;
0347     }
0348 
0349     if (!block->vec_cnt)
0350         goto out;
0351 
0352     atomic_inc(&lc->io_blocks);
0353     bio = bio_alloc(lc->logdev->bdev, bio_max_segs(block->vec_cnt),
0354             REQ_OP_WRITE, GFP_KERNEL);
0355     bio->bi_iter.bi_size = 0;
0356     bio->bi_iter.bi_sector = sector;
0357     bio->bi_end_io = log_end_io;
0358     bio->bi_private = lc;
0359 
0360     for (i = 0; i < block->vec_cnt; i++) {
0361         /*
0362          * The page offset is always 0 because we allocate a new page
0363          * for every bvec in the original bio for simplicity sake.
0364          */
0365         ret = bio_add_page(bio, block->vecs[i].bv_page,
0366                    block->vecs[i].bv_len, 0);
0367         if (ret != block->vecs[i].bv_len) {
0368             atomic_inc(&lc->io_blocks);
0369             submit_bio(bio);
0370             bio = bio_alloc(lc->logdev->bdev,
0371                     bio_max_segs(block->vec_cnt - i),
0372                     REQ_OP_WRITE, GFP_KERNEL);
0373             bio->bi_iter.bi_size = 0;
0374             bio->bi_iter.bi_sector = sector;
0375             bio->bi_end_io = log_end_io;
0376             bio->bi_private = lc;
0377 
0378             ret = bio_add_page(bio, block->vecs[i].bv_page,
0379                        block->vecs[i].bv_len, 0);
0380             if (ret != block->vecs[i].bv_len) {
0381                 DMERR("Couldn't add page on new bio?");
0382                 bio_put(bio);
0383                 goto error;
0384             }
0385         }
0386         sector += block->vecs[i].bv_len >> SECTOR_SHIFT;
0387     }
0388     submit_bio(bio);
0389 out:
0390     kfree(block->data);
0391     kfree(block);
0392     put_pending_block(lc);
0393     return 0;
0394 error:
0395     free_pending_block(lc, block);
0396     put_io_block(lc);
0397     return -1;
0398 }
0399 
0400 static int log_super(struct log_writes_c *lc)
0401 {
0402     struct log_write_super super;
0403 
0404     super.magic = cpu_to_le64(WRITE_LOG_MAGIC);
0405     super.version = cpu_to_le64(WRITE_LOG_VERSION);
0406     super.nr_entries = cpu_to_le64(lc->logged_entries);
0407     super.sectorsize = cpu_to_le32(lc->sectorsize);
0408 
0409     if (write_metadata(lc, &super, sizeof(super), NULL, 0,
0410                WRITE_LOG_SUPER_SECTOR)) {
0411         DMERR("Couldn't write super");
0412         return -1;
0413     }
0414 
0415     /*
0416      * Super sector should be writen in-order, otherwise the
0417      * nr_entries could be rewritten incorrectly by an old bio.
0418      */
0419     wait_for_completion_io(&lc->super_done);
0420 
0421     return 0;
0422 }
0423 
0424 static inline sector_t logdev_last_sector(struct log_writes_c *lc)
0425 {
0426     return bdev_nr_sectors(lc->logdev->bdev);
0427 }
0428 
0429 static int log_writes_kthread(void *arg)
0430 {
0431     struct log_writes_c *lc = (struct log_writes_c *)arg;
0432     sector_t sector = 0;
0433 
0434     while (!kthread_should_stop()) {
0435         bool super = false;
0436         bool logging_enabled;
0437         struct pending_block *block = NULL;
0438         int ret;
0439 
0440         spin_lock_irq(&lc->blocks_lock);
0441         if (!list_empty(&lc->logging_blocks)) {
0442             block = list_first_entry(&lc->logging_blocks,
0443                          struct pending_block, list);
0444             list_del_init(&block->list);
0445             if (!lc->logging_enabled)
0446                 goto next;
0447 
0448             sector = lc->next_sector;
0449             if (!(block->flags & LOG_DISCARD_FLAG))
0450                 lc->next_sector += dev_to_bio_sectors(lc, block->nr_sectors);
0451             lc->next_sector += dev_to_bio_sectors(lc, 1);
0452 
0453             /*
0454              * Apparently the size of the device may not be known
0455              * right away, so handle this properly.
0456              */
0457             if (!lc->end_sector)
0458                 lc->end_sector = logdev_last_sector(lc);
0459             if (lc->end_sector &&
0460                 lc->next_sector >= lc->end_sector) {
0461                 DMERR("Ran out of space on the logdev");
0462                 lc->logging_enabled = false;
0463                 goto next;
0464             }
0465             lc->logged_entries++;
0466             atomic_inc(&lc->io_blocks);
0467 
0468             super = (block->flags & (LOG_FUA_FLAG | LOG_MARK_FLAG));
0469             if (super)
0470                 atomic_inc(&lc->io_blocks);
0471         }
0472 next:
0473         logging_enabled = lc->logging_enabled;
0474         spin_unlock_irq(&lc->blocks_lock);
0475         if (block) {
0476             if (logging_enabled) {
0477                 ret = log_one_block(lc, block, sector);
0478                 if (!ret && super)
0479                     ret = log_super(lc);
0480                 if (ret) {
0481                     spin_lock_irq(&lc->blocks_lock);
0482                     lc->logging_enabled = false;
0483                     spin_unlock_irq(&lc->blocks_lock);
0484                 }
0485             } else
0486                 free_pending_block(lc, block);
0487             continue;
0488         }
0489 
0490         if (!try_to_freeze()) {
0491             set_current_state(TASK_INTERRUPTIBLE);
0492             if (!kthread_should_stop() &&
0493                 list_empty(&lc->logging_blocks))
0494                 schedule();
0495             __set_current_state(TASK_RUNNING);
0496         }
0497     }
0498     return 0;
0499 }
0500 
0501 /*
0502  * Construct a log-writes mapping:
0503  * log-writes <dev_path> <log_dev_path>
0504  */
0505 static int log_writes_ctr(struct dm_target *ti, unsigned int argc, char **argv)
0506 {
0507     struct log_writes_c *lc;
0508     struct dm_arg_set as;
0509     const char *devname, *logdevname;
0510     int ret;
0511 
0512     as.argc = argc;
0513     as.argv = argv;
0514 
0515     if (argc < 2) {
0516         ti->error = "Invalid argument count";
0517         return -EINVAL;
0518     }
0519 
0520     lc = kzalloc(sizeof(struct log_writes_c), GFP_KERNEL);
0521     if (!lc) {
0522         ti->error = "Cannot allocate context";
0523         return -ENOMEM;
0524     }
0525     spin_lock_init(&lc->blocks_lock);
0526     INIT_LIST_HEAD(&lc->unflushed_blocks);
0527     INIT_LIST_HEAD(&lc->logging_blocks);
0528     init_waitqueue_head(&lc->wait);
0529     init_completion(&lc->super_done);
0530     atomic_set(&lc->io_blocks, 0);
0531     atomic_set(&lc->pending_blocks, 0);
0532 
0533     devname = dm_shift_arg(&as);
0534     ret = dm_get_device(ti, devname, dm_table_get_mode(ti->table), &lc->dev);
0535     if (ret) {
0536         ti->error = "Device lookup failed";
0537         goto bad;
0538     }
0539 
0540     logdevname = dm_shift_arg(&as);
0541     ret = dm_get_device(ti, logdevname, dm_table_get_mode(ti->table),
0542                 &lc->logdev);
0543     if (ret) {
0544         ti->error = "Log device lookup failed";
0545         dm_put_device(ti, lc->dev);
0546         goto bad;
0547     }
0548 
0549     lc->sectorsize = bdev_logical_block_size(lc->dev->bdev);
0550     lc->sectorshift = ilog2(lc->sectorsize);
0551     lc->log_kthread = kthread_run(log_writes_kthread, lc, "log-write");
0552     if (IS_ERR(lc->log_kthread)) {
0553         ret = PTR_ERR(lc->log_kthread);
0554         ti->error = "Couldn't alloc kthread";
0555         dm_put_device(ti, lc->dev);
0556         dm_put_device(ti, lc->logdev);
0557         goto bad;
0558     }
0559 
0560     /*
0561      * next_sector is in 512b sectors to correspond to what bi_sector expects.
0562      * The super starts at sector 0, and the next_sector is the next logical
0563      * one based on the sectorsize of the device.
0564      */
0565     lc->next_sector = lc->sectorsize >> SECTOR_SHIFT;
0566     lc->logging_enabled = true;
0567     lc->end_sector = logdev_last_sector(lc);
0568     lc->device_supports_discard = true;
0569 
0570     ti->num_flush_bios = 1;
0571     ti->flush_supported = true;
0572     ti->num_discard_bios = 1;
0573     ti->discards_supported = true;
0574     ti->per_io_data_size = sizeof(struct per_bio_data);
0575     ti->private = lc;
0576     return 0;
0577 
0578 bad:
0579     kfree(lc);
0580     return ret;
0581 }
0582 
0583 static int log_mark(struct log_writes_c *lc, char *data)
0584 {
0585     struct pending_block *block;
0586     size_t maxsize = lc->sectorsize - sizeof(struct log_write_entry);
0587 
0588     block = kzalloc(sizeof(struct pending_block), GFP_KERNEL);
0589     if (!block) {
0590         DMERR("Error allocating pending block");
0591         return -ENOMEM;
0592     }
0593 
0594     block->data = kstrndup(data, maxsize - 1, GFP_KERNEL);
0595     if (!block->data) {
0596         DMERR("Error copying mark data");
0597         kfree(block);
0598         return -ENOMEM;
0599     }
0600     atomic_inc(&lc->pending_blocks);
0601     block->datalen = strlen(block->data);
0602     block->flags |= LOG_MARK_FLAG;
0603     spin_lock_irq(&lc->blocks_lock);
0604     list_add_tail(&block->list, &lc->logging_blocks);
0605     spin_unlock_irq(&lc->blocks_lock);
0606     wake_up_process(lc->log_kthread);
0607     return 0;
0608 }
0609 
0610 static void log_writes_dtr(struct dm_target *ti)
0611 {
0612     struct log_writes_c *lc = ti->private;
0613 
0614     spin_lock_irq(&lc->blocks_lock);
0615     list_splice_init(&lc->unflushed_blocks, &lc->logging_blocks);
0616     spin_unlock_irq(&lc->blocks_lock);
0617 
0618     /*
0619      * This is just nice to have since it'll update the super to include the
0620      * unflushed blocks, if it fails we don't really care.
0621      */
0622     log_mark(lc, "dm-log-writes-end");
0623     wake_up_process(lc->log_kthread);
0624     wait_event(lc->wait, !atomic_read(&lc->io_blocks) &&
0625            !atomic_read(&lc->pending_blocks));
0626     kthread_stop(lc->log_kthread);
0627 
0628     WARN_ON(!list_empty(&lc->logging_blocks));
0629     WARN_ON(!list_empty(&lc->unflushed_blocks));
0630     dm_put_device(ti, lc->dev);
0631     dm_put_device(ti, lc->logdev);
0632     kfree(lc);
0633 }
0634 
0635 static void normal_map_bio(struct dm_target *ti, struct bio *bio)
0636 {
0637     struct log_writes_c *lc = ti->private;
0638 
0639     bio_set_dev(bio, lc->dev->bdev);
0640 }
0641 
0642 static int log_writes_map(struct dm_target *ti, struct bio *bio)
0643 {
0644     struct log_writes_c *lc = ti->private;
0645     struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
0646     struct pending_block *block;
0647     struct bvec_iter iter;
0648     struct bio_vec bv;
0649     size_t alloc_size;
0650     int i = 0;
0651     bool flush_bio = (bio->bi_opf & REQ_PREFLUSH);
0652     bool fua_bio = (bio->bi_opf & REQ_FUA);
0653     bool discard_bio = (bio_op(bio) == REQ_OP_DISCARD);
0654     bool meta_bio = (bio->bi_opf & REQ_META);
0655 
0656     pb->block = NULL;
0657 
0658     /* Don't bother doing anything if logging has been disabled */
0659     if (!lc->logging_enabled)
0660         goto map_bio;
0661 
0662     /*
0663      * Map reads as normal.
0664      */
0665     if (bio_data_dir(bio) == READ)
0666         goto map_bio;
0667 
0668     /* No sectors and not a flush?  Don't care */
0669     if (!bio_sectors(bio) && !flush_bio)
0670         goto map_bio;
0671 
0672     /*
0673      * Discards will have bi_size set but there's no actual data, so just
0674      * allocate the size of the pending block.
0675      */
0676     if (discard_bio)
0677         alloc_size = sizeof(struct pending_block);
0678     else
0679         alloc_size = struct_size(block, vecs, bio_segments(bio));
0680 
0681     block = kzalloc(alloc_size, GFP_NOIO);
0682     if (!block) {
0683         DMERR("Error allocating pending block");
0684         spin_lock_irq(&lc->blocks_lock);
0685         lc->logging_enabled = false;
0686         spin_unlock_irq(&lc->blocks_lock);
0687         return DM_MAPIO_KILL;
0688     }
0689     INIT_LIST_HEAD(&block->list);
0690     pb->block = block;
0691     atomic_inc(&lc->pending_blocks);
0692 
0693     if (flush_bio)
0694         block->flags |= LOG_FLUSH_FLAG;
0695     if (fua_bio)
0696         block->flags |= LOG_FUA_FLAG;
0697     if (discard_bio)
0698         block->flags |= LOG_DISCARD_FLAG;
0699     if (meta_bio)
0700         block->flags |= LOG_METADATA_FLAG;
0701 
0702     block->sector = bio_to_dev_sectors(lc, bio->bi_iter.bi_sector);
0703     block->nr_sectors = bio_to_dev_sectors(lc, bio_sectors(bio));
0704 
0705     /* We don't need the data, just submit */
0706     if (discard_bio) {
0707         WARN_ON(flush_bio || fua_bio);
0708         if (lc->device_supports_discard)
0709             goto map_bio;
0710         bio_endio(bio);
0711         return DM_MAPIO_SUBMITTED;
0712     }
0713 
0714     /* Flush bio, splice the unflushed blocks onto this list and submit */
0715     if (flush_bio && !bio_sectors(bio)) {
0716         spin_lock_irq(&lc->blocks_lock);
0717         list_splice_init(&lc->unflushed_blocks, &block->list);
0718         spin_unlock_irq(&lc->blocks_lock);
0719         goto map_bio;
0720     }
0721 
0722     /*
0723      * We will write this bio somewhere else way later so we need to copy
0724      * the actual contents into new pages so we know the data will always be
0725      * there.
0726      *
0727      * We do this because this could be a bio from O_DIRECT in which case we
0728      * can't just hold onto the page until some later point, we have to
0729      * manually copy the contents.
0730      */
0731     bio_for_each_segment(bv, bio, iter) {
0732         struct page *page;
0733         void *dst;
0734 
0735         page = alloc_page(GFP_NOIO);
0736         if (!page) {
0737             DMERR("Error allocing page");
0738             free_pending_block(lc, block);
0739             spin_lock_irq(&lc->blocks_lock);
0740             lc->logging_enabled = false;
0741             spin_unlock_irq(&lc->blocks_lock);
0742             return DM_MAPIO_KILL;
0743         }
0744 
0745         dst = kmap_atomic(page);
0746         memcpy_from_bvec(dst, &bv);
0747         kunmap_atomic(dst);
0748         block->vecs[i].bv_page = page;
0749         block->vecs[i].bv_len = bv.bv_len;
0750         block->vec_cnt++;
0751         i++;
0752     }
0753 
0754     /* Had a flush with data in it, weird */
0755     if (flush_bio) {
0756         spin_lock_irq(&lc->blocks_lock);
0757         list_splice_init(&lc->unflushed_blocks, &block->list);
0758         spin_unlock_irq(&lc->blocks_lock);
0759     }
0760 map_bio:
0761     normal_map_bio(ti, bio);
0762     return DM_MAPIO_REMAPPED;
0763 }
0764 
0765 static int normal_end_io(struct dm_target *ti, struct bio *bio,
0766         blk_status_t *error)
0767 {
0768     struct log_writes_c *lc = ti->private;
0769     struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data));
0770 
0771     if (bio_data_dir(bio) == WRITE && pb->block) {
0772         struct pending_block *block = pb->block;
0773         unsigned long flags;
0774 
0775         spin_lock_irqsave(&lc->blocks_lock, flags);
0776         if (block->flags & LOG_FLUSH_FLAG) {
0777             list_splice_tail_init(&block->list, &lc->logging_blocks);
0778             list_add_tail(&block->list, &lc->logging_blocks);
0779             wake_up_process(lc->log_kthread);
0780         } else if (block->flags & LOG_FUA_FLAG) {
0781             list_add_tail(&block->list, &lc->logging_blocks);
0782             wake_up_process(lc->log_kthread);
0783         } else
0784             list_add_tail(&block->list, &lc->unflushed_blocks);
0785         spin_unlock_irqrestore(&lc->blocks_lock, flags);
0786     }
0787 
0788     return DM_ENDIO_DONE;
0789 }
0790 
0791 /*
0792  * INFO format: <logged entries> <highest allocated sector>
0793  */
0794 static void log_writes_status(struct dm_target *ti, status_type_t type,
0795                   unsigned status_flags, char *result,
0796                   unsigned maxlen)
0797 {
0798     unsigned sz = 0;
0799     struct log_writes_c *lc = ti->private;
0800 
0801     switch (type) {
0802     case STATUSTYPE_INFO:
0803         DMEMIT("%llu %llu", lc->logged_entries,
0804                (unsigned long long)lc->next_sector - 1);
0805         if (!lc->logging_enabled)
0806             DMEMIT(" logging_disabled");
0807         break;
0808 
0809     case STATUSTYPE_TABLE:
0810         DMEMIT("%s %s", lc->dev->name, lc->logdev->name);
0811         break;
0812 
0813     case STATUSTYPE_IMA:
0814         *result = '\0';
0815         break;
0816     }
0817 }
0818 
0819 static int log_writes_prepare_ioctl(struct dm_target *ti,
0820                     struct block_device **bdev)
0821 {
0822     struct log_writes_c *lc = ti->private;
0823     struct dm_dev *dev = lc->dev;
0824 
0825     *bdev = dev->bdev;
0826     /*
0827      * Only pass ioctls through if the device sizes match exactly.
0828      */
0829     if (ti->len != bdev_nr_sectors(dev->bdev))
0830         return 1;
0831     return 0;
0832 }
0833 
0834 static int log_writes_iterate_devices(struct dm_target *ti,
0835                       iterate_devices_callout_fn fn,
0836                       void *data)
0837 {
0838     struct log_writes_c *lc = ti->private;
0839 
0840     return fn(ti, lc->dev, 0, ti->len, data);
0841 }
0842 
0843 /*
0844  * Messages supported:
0845  *   mark <mark data> - specify the marked data.
0846  */
0847 static int log_writes_message(struct dm_target *ti, unsigned argc, char **argv,
0848                   char *result, unsigned maxlen)
0849 {
0850     int r = -EINVAL;
0851     struct log_writes_c *lc = ti->private;
0852 
0853     if (argc != 2) {
0854         DMWARN("Invalid log-writes message arguments, expect 2 arguments, got %d", argc);
0855         return r;
0856     }
0857 
0858     if (!strcasecmp(argv[0], "mark"))
0859         r = log_mark(lc, argv[1]);
0860     else
0861         DMWARN("Unrecognised log writes target message received: %s", argv[0]);
0862 
0863     return r;
0864 }
0865 
0866 static void log_writes_io_hints(struct dm_target *ti, struct queue_limits *limits)
0867 {
0868     struct log_writes_c *lc = ti->private;
0869 
0870     if (!bdev_max_discard_sectors(lc->dev->bdev)) {
0871         lc->device_supports_discard = false;
0872         limits->discard_granularity = lc->sectorsize;
0873         limits->max_discard_sectors = (UINT_MAX >> SECTOR_SHIFT);
0874     }
0875     limits->logical_block_size = bdev_logical_block_size(lc->dev->bdev);
0876     limits->physical_block_size = bdev_physical_block_size(lc->dev->bdev);
0877     limits->io_min = limits->physical_block_size;
0878 }
0879 
0880 #if IS_ENABLED(CONFIG_FS_DAX)
0881 static struct dax_device *log_writes_dax_pgoff(struct dm_target *ti,
0882         pgoff_t *pgoff)
0883 {
0884     struct log_writes_c *lc = ti->private;
0885 
0886     *pgoff += (get_start_sect(lc->dev->bdev) >> PAGE_SECTORS_SHIFT);
0887     return lc->dev->dax_dev;
0888 }
0889 
0890 static long log_writes_dax_direct_access(struct dm_target *ti, pgoff_t pgoff,
0891         long nr_pages, enum dax_access_mode mode, void **kaddr,
0892         pfn_t *pfn)
0893 {
0894     struct dax_device *dax_dev = log_writes_dax_pgoff(ti, &pgoff);
0895 
0896     return dax_direct_access(dax_dev, pgoff, nr_pages, mode, kaddr, pfn);
0897 }
0898 
0899 static int log_writes_dax_zero_page_range(struct dm_target *ti, pgoff_t pgoff,
0900                       size_t nr_pages)
0901 {
0902     struct dax_device *dax_dev = log_writes_dax_pgoff(ti, &pgoff);
0903 
0904     return dax_zero_page_range(dax_dev, pgoff, nr_pages << PAGE_SHIFT);
0905 }
0906 
0907 static size_t log_writes_dax_recovery_write(struct dm_target *ti,
0908         pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i)
0909 {
0910     struct dax_device *dax_dev = log_writes_dax_pgoff(ti, &pgoff);
0911 
0912     return dax_recovery_write(dax_dev, pgoff, addr, bytes, i);
0913 }
0914 
0915 #else
0916 #define log_writes_dax_direct_access NULL
0917 #define log_writes_dax_zero_page_range NULL
0918 #define log_writes_dax_recovery_write NULL
0919 #endif
0920 
0921 static struct target_type log_writes_target = {
0922     .name   = "log-writes",
0923     .version = {1, 1, 0},
0924     .module = THIS_MODULE,
0925     .ctr    = log_writes_ctr,
0926     .dtr    = log_writes_dtr,
0927     .map    = log_writes_map,
0928     .end_io = normal_end_io,
0929     .status = log_writes_status,
0930     .prepare_ioctl = log_writes_prepare_ioctl,
0931     .message = log_writes_message,
0932     .iterate_devices = log_writes_iterate_devices,
0933     .io_hints = log_writes_io_hints,
0934     .direct_access = log_writes_dax_direct_access,
0935     .dax_zero_page_range = log_writes_dax_zero_page_range,
0936     .dax_recovery_write = log_writes_dax_recovery_write,
0937 };
0938 
0939 static int __init dm_log_writes_init(void)
0940 {
0941     int r = dm_register_target(&log_writes_target);
0942 
0943     if (r < 0)
0944         DMERR("register failed %d", r);
0945 
0946     return r;
0947 }
0948 
0949 static void __exit dm_log_writes_exit(void)
0950 {
0951     dm_unregister_target(&log_writes_target);
0952 }
0953 
0954 module_init(dm_log_writes_init);
0955 module_exit(dm_log_writes_exit);
0956 
0957 MODULE_DESCRIPTION(DM_NAME " log writes target");
0958 MODULE_AUTHOR("Josef Bacik <jbacik@fb.com>");
0959 MODULE_LICENSE("GPL");
[ 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!