Whamcloud - gitweb
LU-4712 llite: lock the inode to be migrated
[fs/lustre-release.git] / lustre / llite / lloop.c
index f94a7ca..f3bf6de 100644 (file)
@@ -27,7 +27,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2014, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #include <asm/uaccess.h>
 
 #include <lustre_lib.h>
-#include <lustre_lite.h>
 #include "llite_internal.h"
 
 #define LLOOP_MAX_SEGMENTS        LNET_MAX_IOV
@@ -118,19 +117,16 @@ enum {
 };
 
 struct lloop_device {
-        int                  lo_number;
-        int                  lo_refcnt;
-        loff_t               lo_offset;
-        loff_t               lo_sizelimit;
-        int                  lo_flags;
-        int                (*ioctl)(struct lloop_device *, int cmd,
-                                    unsigned long arg);
+       int                  lo_number;
+       int                  lo_refcnt;
+       loff_t               lo_offset;
+       loff_t               lo_sizelimit;
+       int                  lo_flags;
+       struct file         *lo_backing_file;
+       struct block_device *lo_device;
+       unsigned             lo_blocksize;
 
-        struct file         *lo_backing_file;
-        struct block_device *lo_device;
-        unsigned             lo_blocksize;
-
-        int                  old_gfp_mask;
+       gfp_t                   old_gfp_mask;
 
        spinlock_t              lo_lock;
        struct bio              *lo_bio;
@@ -138,20 +134,20 @@ struct lloop_device {
        int                     lo_state;
        struct semaphore        lo_sem;
        struct mutex            lo_ctl_mutex;
-        cfs_atomic_t         lo_pending;
-        cfs_waitq_t          lo_bh_wait;
+       atomic_t                lo_pending;
+       wait_queue_head_t       lo_bh_wait;
 
-        struct request_queue *lo_queue;
+       struct request_queue *lo_queue;
 
-        const struct lu_env *lo_env;
-        struct cl_io         lo_io;
-        struct ll_dio_pages  lo_pvec;
+       const struct lu_env *lo_env;
+       struct cl_io         lo_io;
+       struct ll_dio_pages  lo_pvec;
 
-        /* data to handle bio for lustre. */
-        struct lo_request_data {
-                struct page *lrd_pages[LLOOP_MAX_SEGMENTS];
-                loff_t       lrd_offsets[LLOOP_MAX_SEGMENTS];
-        } lo_requests[1];
+       /* data to handle bio for lustre. */
+       struct lo_request_data {
+               struct page *lrd_pages[LLOOP_MAX_SEGMENTS];
+               loff_t       lrd_offsets[LLOOP_MAX_SEGMENTS];
+       } lo_requests[1];
 };
 
 /*
@@ -195,10 +191,15 @@ static int do_bio_lustrebacked(struct lloop_device *lo, struct bio *head)
         struct cl_object     *obj = ll_i2info(inode)->lli_clob;
         pgoff_t               offset;
         int                   ret;
-        int                   i;
+#ifdef HAVE_BVEC_ITER
+       struct bvec_iter      iter;
+       struct bio_vec        bvec;
+#else
+       int                   iter;
+       struct bio_vec       *bvec;
+#endif
         int                   rw;
-        obd_count             page_count = 0;
-        struct bio_vec       *bvec;
+       size_t                page_count = 0;
         struct bio           *bio;
         ssize_t               bytes;
 
@@ -221,18 +222,30 @@ static int do_bio_lustrebacked(struct lloop_device *lo, struct bio *head)
         for (bio = head; bio != NULL; bio = bio->bi_next) {
                 LASSERT(rw == bio->bi_rw);
 
-                offset = (pgoff_t)(bio->bi_sector << 9) + lo->lo_offset;
-                bio_for_each_segment(bvec, bio, i) {
-                        BUG_ON(bvec->bv_offset != 0);
-                        BUG_ON(bvec->bv_len != CFS_PAGE_SIZE);
+#ifdef HAVE_BVEC_ITER
+               offset = (pgoff_t)(bio->bi_iter.bi_sector << 9) + lo->lo_offset;
+               bio_for_each_segment(bvec, bio, iter) {
+                       BUG_ON(bvec.bv_offset != 0);
+                       BUG_ON(bvec.bv_len != PAGE_CACHE_SIZE);
 
-                        pages[page_count] = bvec->bv_page;
-                        offsets[page_count] = offset;
-                        page_count++;
-                        offset += bvec->bv_len;
-                }
-                LASSERT(page_count <= LLOOP_MAX_SEGMENTS);
-        }
+                       pages[page_count] = bvec.bv_page;
+                       offsets[page_count] = offset;
+                       page_count++;
+                       offset += bvec.bv_len;
+#else
+               offset = (pgoff_t)(bio->bi_sector << 9) + lo->lo_offset;
+               bio_for_each_segment(bvec, bio, iter) {
+                       BUG_ON(bvec->bv_offset != 0);
+                       BUG_ON(bvec->bv_len != PAGE_CACHE_SIZE);
+
+                       pages[page_count] = bvec->bv_page;
+                       offsets[page_count] = offset;
+                       page_count++;
+                       offset += bvec->bv_len;
+#endif
+               }
+               LASSERT(page_count <= LLOOP_MAX_SEGMENTS);
+       }
 
         ll_stats_ops_tally(ll_i2sbi(inode),
                         (rw == WRITE) ? LPROC_LL_BRW_WRITE : LPROC_LL_BRW_READ,
@@ -284,9 +297,9 @@ static void loop_add_bio(struct lloop_device *lo, struct bio *bio)
                lo->lo_bio = lo->lo_biotail = bio;
        spin_unlock_irqrestore(&lo->lo_lock, flags);
 
-       cfs_atomic_inc(&lo->lo_pending);
-       if (cfs_waitq_active(&lo->lo_bh_wait))
-               cfs_waitq_signal(&lo->lo_bh_wait);
+       atomic_inc(&lo->lo_pending);
+       if (waitqueue_active(&lo->lo_bh_wait))
+               wake_up(&lo->lo_bh_wait);
 }
 
 /*
@@ -313,12 +326,17 @@ static unsigned int loop_get_bio(struct lloop_device *lo, struct bio **req)
         rw = first->bi_rw;
         bio = &lo->lo_bio;
         while (*bio && (*bio)->bi_rw == rw) {
-                CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n",
-                       (unsigned long long)(*bio)->bi_sector, (*bio)->bi_size,
-                       page_count, (*bio)->bi_vcnt);
-                if (page_count + (*bio)->bi_vcnt > LLOOP_MAX_SEGMENTS)
-                        break;
-
+#ifdef HAVE_BVEC_ITER
+               CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n",
+                      (unsigned long long)(*bio)->bi_iter.bi_sector,
+                      (*bio)->bi_iter.bi_size, page_count, (*bio)->bi_vcnt);
+#else
+               CDEBUG(D_INFO, "bio sector %llu size %u count %u vcnt%u \n",
+                      (unsigned long long)(*bio)->bi_sector, (*bio)->bi_size,
+                      page_count, (*bio)->bi_vcnt);
+#endif
+               if (page_count + (*bio)->bi_vcnt > LLOOP_MAX_SEGMENTS)
+                       break;
 
                 page_count += (*bio)->bi_vcnt;
                 count++;
@@ -348,8 +366,14 @@ loop_make_request(struct request_queue *q, struct bio *old_bio)
         if (!lo)
                 goto err;
 
-        CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
-               (unsigned long long)old_bio->bi_sector, old_bio->bi_size);
+#ifdef HAVE_BVEC_ITER
+       CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
+              (unsigned long long)old_bio->bi_iter.bi_sector,
+              old_bio->bi_iter.bi_size);
+#else
+       CDEBUG(D_INFO, "submit bio sector %llu size %u\n",
+              (unsigned long long)old_bio->bi_sector, old_bio->bi_size);
+#endif
 
        spin_lock_irq(&lo->lo_lock);
        inactive = (lo->lo_state != LLOOP_BOUND);
@@ -369,7 +393,7 @@ loop_make_request(struct request_queue *q, struct bio *old_bio)
         loop_add_bio(lo, old_bio);
        LL_MRF_RETURN(0);
 err:
-       cfs_bio_io_error(old_bio, old_bio->bi_size);
+       bio_io_error(old_bio);
        LL_MRF_RETURN(0);
 }
 
@@ -393,15 +417,15 @@ static inline void loop_handle_bio(struct lloop_device *lo, struct bio *bio)
         while (bio) {
                 struct bio *tmp = bio->bi_next;
                 bio->bi_next = NULL;
-                cfs_bio_endio(bio, bio->bi_size, ret);
+               bio_endio(bio, ret);
                 bio = tmp;
         }
 }
 
 static inline int loop_active(struct lloop_device *lo)
 {
-        return cfs_atomic_read(&lo->lo_pending) ||
-                (lo->lo_state == LLOOP_RUNDOWN);
+       return atomic_read(&lo->lo_pending) ||
+              (lo->lo_state == LLOOP_RUNDOWN);
 }
 
 /*
@@ -420,8 +444,6 @@ static int loop_thread(void *data)
         int refcheck;
         int ret = 0;
 
-        daemonize("lloop%d", lo->lo_number);
-
         set_user_nice(current, -20);
 
         lo->lo_state = LLOOP_BOUND;
@@ -441,8 +463,8 @@ static int loop_thread(void *data)
        up(&lo->lo_sem);
 
        for (;;) {
-               cfs_wait_event(lo->lo_bh_wait, loop_active(lo));
-               if (!cfs_atomic_read(&lo->lo_pending)) {
+               wait_event(lo->lo_bh_wait, loop_active(lo));
+               if (!atomic_read(&lo->lo_pending)) {
                        int exiting = 0;
                        spin_lock_irq(&lo->lo_lock);
                        exiting = (lo->lo_state == LLOOP_RUNDOWN);
@@ -465,21 +487,21 @@ static int loop_thread(void *data)
                 } else {
                         times++;
                 }
-                if ((times & 127) == 0) {
-                        CDEBUG(D_INFO, "total: %lu, count: %lu, avg: %lu\n",
-                               total_count, times, total_count / times);
-                }
-
-                LASSERT(bio != NULL);
-                LASSERT(count <= cfs_atomic_read(&lo->lo_pending));
-                loop_handle_bio(lo, bio);
-                cfs_atomic_sub(count, &lo->lo_pending);
-        }
-        cl_env_put(env, &refcheck);
+               if ((times & 127) == 0) {
+                       CDEBUG(D_INFO, "total: %lu, count: %lu, avg: %lu\n",
+                              total_count, times, total_count / times);
+               }
+
+               LASSERT(bio != NULL);
+               LASSERT(count <= atomic_read(&lo->lo_pending));
+               loop_handle_bio(lo, bio);
+               atomic_sub(count, &lo->lo_pending);
+       }
+       cl_env_put(env, &refcheck);
 
 out:
        up(&lo->lo_sem);
-        return ret;
+       return ret;
 }
 
 static int loop_set_fd(struct lloop_device *lo, struct file *unused,
@@ -491,8 +513,8 @@ static int loop_set_fd(struct lloop_device *lo, struct file *unused,
         int                   error;
         loff_t                size;
 
-        if (!cfs_try_module_get(THIS_MODULE))
-                return -ENODEV;
+       if (!try_module_get(THIS_MODULE))
+               return -ENODEV;
 
         error = -EBUSY;
         if (lo->lo_state != LLOOP_UNBOUND)
@@ -520,11 +542,10 @@ static int loop_set_fd(struct lloop_device *lo, struct file *unused,
 
         set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
 
-        lo->lo_blocksize = CFS_PAGE_SIZE;
+       lo->lo_blocksize = PAGE_CACHE_SIZE;
         lo->lo_device = bdev;
         lo->lo_flags = lo_flags;
         lo->lo_backing_file = file;
-        lo->ioctl = NULL;
         lo->lo_sizelimit = 0;
         lo->old_gfp_mask = mapping_gfp_mask(mapping);
         mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
@@ -541,34 +562,31 @@ static int loop_set_fd(struct lloop_device *lo, struct file *unused,
         lo->lo_queue->unplug_fn = loop_unplug;
 #endif
 
-        /* queue parameters */
-        CLASSERT(CFS_PAGE_SIZE < (1 << (sizeof(unsigned short) * 8)));
-        blk_queue_logical_block_size(lo->lo_queue,
-                                     (unsigned short)CFS_PAGE_SIZE);
-        blk_queue_max_hw_sectors(lo->lo_queue,
-                                 LLOOP_MAX_SEGMENTS << (CFS_PAGE_SHIFT - 9));
-        blk_queue_max_segments(lo->lo_queue, LLOOP_MAX_SEGMENTS);
+       /* queue parameters */
+       blk_queue_max_hw_sectors(lo->lo_queue,
+                                LLOOP_MAX_SEGMENTS << (PAGE_CACHE_SHIFT - 9));
+       blk_queue_max_segments(lo->lo_queue, LLOOP_MAX_SEGMENTS);
 
         set_capacity(disks[lo->lo_number], size);
         bd_set_size(bdev, size << 9);
 
-        set_blocksize(bdev, lo->lo_blocksize);
+       set_blocksize(bdev, lo->lo_blocksize);
 
-        cfs_create_thread(loop_thread, lo, CLONE_KERNEL);
+       kthread_run(loop_thread, lo, "lloop%d", lo->lo_number);
        down(&lo->lo_sem);
-        return 0;
+       return 0;
 
- out:
-        /* This is safe: open() is still holding a reference. */
-        cfs_module_put(THIS_MODULE);
-        return error;
+out:
+       /* This is safe: open() is still holding a reference. */
+       module_put(THIS_MODULE);
+       return error;
 }
 
 static int loop_clr_fd(struct lloop_device *lo, struct block_device *bdev,
                        int count)
 {
         struct file *filp = lo->lo_backing_file;
-        int gfp = lo->old_gfp_mask;
+       gfp_t gfp = lo->old_gfp_mask;
 
         if (lo->lo_state != LLOOP_BOUND)
                 return -ENXIO;
@@ -582,35 +600,28 @@ static int loop_clr_fd(struct lloop_device *lo, struct block_device *bdev,
        spin_lock_irq(&lo->lo_lock);
        lo->lo_state = LLOOP_RUNDOWN;
        spin_unlock_irq(&lo->lo_lock);
-       cfs_waitq_signal(&lo->lo_bh_wait);
+       wake_up(&lo->lo_bh_wait);
 
        down(&lo->lo_sem);
         lo->lo_backing_file = NULL;
-        lo->ioctl = NULL;
         lo->lo_device = NULL;
         lo->lo_offset = 0;
         lo->lo_sizelimit = 0;
         lo->lo_flags = 0;
-        ll_invalidate_bdev(bdev, 0);
+       invalidate_bdev(bdev);
         set_capacity(disks[lo->lo_number], 0);
         bd_set_size(bdev, 0);
         mapping_set_gfp_mask(filp->f_mapping, gfp);
         lo->lo_state = LLOOP_UNBOUND;
-        fput(filp);
-        /* This is safe: open() is still holding a reference. */
-        cfs_module_put(THIS_MODULE);
-        return 0;
+       fput(filp);
+       /* This is safe: open() is still holding a reference. */
+       module_put(THIS_MODULE);
+       return 0;
 }
 
-#ifdef HAVE_BLKDEV_PUT_2ARGS
 static int lo_open(struct block_device *bdev, fmode_t mode)
 {
         struct lloop_device *lo = bdev->bd_disk->private_data;
-#else
-static int lo_open(struct inode *inode, struct file *file)
-{
-        struct lloop_device *lo = inode->i_bdev->bd_disk->private_data;
-#endif
 
        mutex_lock(&lo->lo_ctl_mutex);
         lo->lo_refcnt++;
@@ -619,53 +630,47 @@ static int lo_open(struct inode *inode, struct file *file)
         return 0;
 }
 
-#ifdef HAVE_BLKDEV_PUT_2ARGS
-static int lo_release(struct gendisk *disk, fmode_t mode)
-{
-        struct lloop_device *lo = disk->private_data;
+#ifdef HAVE_BLKDEV_RELEASE_RETURN_INT
+static int
 #else
-static int lo_release(struct inode *inode, struct file *file)
-{
-        struct lloop_device *lo = inode->i_bdev->bd_disk->private_data;
+static void
 #endif
+lo_release(struct gendisk *disk, fmode_t mode)
+{
+       struct lloop_device *lo = disk->private_data;
 
        mutex_lock(&lo->lo_ctl_mutex);
-        --lo->lo_refcnt;
+       --lo->lo_refcnt;
        mutex_unlock(&lo->lo_ctl_mutex);
-
-        return 0;
+#ifdef HAVE_BLKDEV_RELEASE_RETURN_INT
+       return 0;
+#endif
 }
 
 /* lloop device node's ioctl function. */
-#ifdef HAVE_BLKDEV_PUT_2ARGS
 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
                     unsigned int cmd, unsigned long arg)
 {
         struct lloop_device *lo = bdev->bd_disk->private_data;
         struct inode *inode = NULL;
         int err = 0;
-#else
-static int lo_ioctl(struct inode *inode, struct file *unused,
-                    unsigned int cmd, unsigned long arg)
-{
-        struct lloop_device *lo = inode->i_bdev->bd_disk->private_data;
-        struct block_device *bdev = inode->i_bdev;
-        int err = 0;
-#endif
 
        mutex_lock(&lloop_mutex);
         switch (cmd) {
         case LL_IOC_LLOOP_DETACH: {
                 err = loop_clr_fd(lo, bdev, 2);
                 if (err == 0)
-                        ll_blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
+                       blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
                 break;
         }
 
         case LL_IOC_LLOOP_INFO: {
                 struct lu_fid fid;
 
-                LASSERT(lo->lo_backing_file != NULL);
+               if (lo->lo_backing_file == NULL) {
+                       err = -ENOENT;
+                       break;
+               }
                 if (inode == NULL)
                         inode = lo->lo_backing_file->f_dentry->d_inode;
                 if (lo->lo_state == LLOOP_BOUND)
@@ -673,9 +678,10 @@ static int lo_ioctl(struct inode *inode, struct file *unused,
                 else
                         fid_zero(&fid);
 
-                if (copy_to_user((struct lu_fid *)arg, &fid, sizeof(fid)))
-                        err = -EFAULT;
-                break;
+               if (copy_to_user((struct lu_fid __user *)arg,
+                                &fid, sizeof(fid)))
+                       err = -EFAULT;
+               break;
         }
 
         default:
@@ -742,8 +748,8 @@ static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file,
                 dev = MKDEV(lloop_major, lo->lo_number);
 
                 /* quit if the used pointer is writable */
-                if (put_user((long)old_encode_dev(dev), (long*)arg))
-                        GOTO(out, err = -EFAULT);
+               if (put_user((long)old_encode_dev(dev), (long __user *)arg))
+                       GOTO(out, err = -EFAULT);
 
                 bdev = blkdev_get_by_dev(dev, file->f_mode, NULL);
                 if (IS_ERR(bdev))
@@ -753,7 +759,7 @@ static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file,
                 err = loop_set_fd(lo, NULL, bdev, file);
                 if (err) {
                         fput(file);
-                        ll_blkdev_put(bdev, 0);
+                       blkdev_put(bdev, 0);
                 }
 
                 break;
@@ -777,7 +783,7 @@ static enum llioc_iter lloop_ioctl(struct inode *unused, struct file *file,
                 bdev = lo->lo_device;
                 err = loop_clr_fd(lo, bdev, 1);
                 if (err == 0)
-                        ll_blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
+                       blkdev_put(bdev, 0); /* grabbed in LLOOP_ATTACH */
 
                 break;
         }
@@ -846,7 +852,7 @@ static int __init lloop_init(void)
 
                mutex_init(&lo->lo_ctl_mutex);
                sema_init(&lo->lo_sem, 0);
-               cfs_waitq_init(&lo->lo_bh_wait);
+               init_waitqueue_head(&lo->lo_bh_wait);
                lo->lo_number = i;
                spin_lock_init(&lo->lo_lock);
                 disk->major = lloop_major;
@@ -889,10 +895,7 @@ static void lloop_exit(void)
                 blk_cleanup_queue(loop_dev[i].lo_queue);
                 put_disk(disks[i]);
         }
-        if (ll_unregister_blkdev(lloop_major, "lloop"))
-                CWARN("lloop: cannot unregister blkdev\n");
-        else
-                CDEBUG(D_CONFIG, "unregistered lloop major %d\n", lloop_major);
+       unregister_blkdev(lloop_major, "lloop");
 
         OBD_FREE(disks, max_loop * sizeof(*disks));
         OBD_FREE(loop_dev, max_loop * sizeof(*loop_dev));